Using instrumental variables to estimate a Cox’s proportional hazards regression subject to additive confounding

The estimation of treatment effects is one of the primary goals of statistics in medicine. Estimation based on observational studies is subject to confounding. Statistical methods for controlling bias due to confounding include regression adjustment, propensity scores and inverse probability weighted estimators. These methods require that all confounders are recorded in the data. The method of instrumental variables (IVs) can eliminate bias in observational studies even in the absence of information on confounders. We propose a method for integrating IVs within the framework of Cox’s proportional hazards model and demonstrate the conditions under which it recovers the causal effect of treatment. The methodology is based on the approximate orthogonality of an instrument with unobserved confounders among those at risk. We derive an estimator as the solution to an estimating equation that resembles the score equation of the partial likelihood in much the same way as the traditional IV estimator resembles the normal equations. To justify this IV estimator for a Cox model we perform simulations to evaluate its operating characteristics. Finally, we apply the estimator to an observational study of the effect of coronary catheterization on survival.     

“Dose-dependent” Impact of Recurrent Cardiac Events on Mortality in Patients with Heart Failure

Background

The mortality impact of recurrent cardiac hospitalizations has not been delineated in community-based heart failure patients. We determined if a “dose-dependent” relationship exists between heart failure events and death, accounting for temporal changes in age, comorbidities, and disease severity.

Methods

Among heart failure patients in the Enhanced Feedback For Effective Cardiac Treatment Study with onset between April 1999 and March 2001, we compared long-term survival (until March 2006) in those with recurrent heart failure or cardiovascular events, relative to those free of such events.

Results

In 9138 patients, 28,442 person-years of follow-up were examined (mean age: 75.3 years, 49.6% male). Recurrent heart failure events occurred 1, 2, 3, and ≥4 times in 2352 (25.7%), 1020 (11.2%), 505 (5.5%), and 596 (6.5%) patients, respectively. Cardiovascular readmissions occurred 1, 2, 3, and ≥4 times in 2522 (27.6%), 1509 (16.5%), 975 (10.7%), and 1672 (18.3%) patients, respectively. Compared with those without recurrent heart failure events, the adjusted relative mortality rates for 1, 2, 3, and ≥4 heart failure events were 2.41 (95% confidence interval [CI], 2.24-2.60), 3.00 (95% CI 2.72-3.32), 4.00 (95% CI, 3.51-4.56), and 5.16 (95% CI, 4.55-5.85), respectively. Compared with those without cardiovascular events, the adjusted relative mortality rates for 1, 2, 3, and ≥4 cardiovascular events were 3.33 (95% CI, 3.05-3.63), 4.61 (95% CI, 4.16-5.10), 6.29 (95% CI, 5.59-7.07), and 8.95 (95% CI, 8.05-9.95), respectively.

Conclusions

The risk of death increases progressively and independently with each heart failure or cardiovascular event. The number of prior events predicts mortality and should be ascertained in patients with heart failure.     

Factors associated with incomplete colonoscopy: a population-based study

BACKGROUND & AIMS: The U.S. Multi-Society Task Force on Colorectal Cancer sets a target of cecal intubation in at least 90% of colonoscopies. We conducted a population-based study to determine the colonoscopy completion rate and to identify factors associated with incomplete procedures. METHODS: Men and women 50 to 74 years of age who underwent a colonoscopy in Ontario between January 1, 1999, and December 31, 2003, were identified. The first (index) colonoscopy was classified as complete or incomplete. A generalized estimating equations model was used to evaluate the association between patient, endoscopist (specialty, colonoscopy volume), and setting (academic hospital, community hospital, private office) factors and incomplete colonoscopy. RESULTS: A total of 331,608 individuals had an index colonoscopy, of which 43,483 (13.1%) were incomplete. Patients with an incomplete colonoscopy were older (odds ratio [OR] 1.20 per 10-year increment; 95% confidence interval [CI]=1.18-1.22), more likely to be female (OR 1.35; 95% CI: 1.30-1.39), have a history of prior abdominal surgery (OR 1.07; 95% CI: 1.05-1.09) or prior pelvic surgery (OR 1.04; 95% CI: 1.01-1.06). For colonoscopies done in a private office, the odds of an incomplete procedure were more than 3-fold greater than for procedures done in an academic hospital (OR 3.57; 95% CI: 2.55-4.98). CONCLUSIONS: In usual clinical practice in Ontario, 13.1% of colonoscopies are incomplete. The factors most strongly associated with incomplete colonoscopy were increased patient age, female sex, and having the procedure in a private office. Quality improvement programs are needed to improve colonoscopy completion rates.     

Screening,Treatment, and Prostate Cancer Mortality in the Seattle Area and Connecticut: Fifteen-year Follow-up

Background  Prior to introduction of the prostate-specific antigen (PSA) test, the Seattle–Puget Sound and Connecticut Surveillance, Epidemiology and End Results (SEER) areas had similar prostate cancer mortality rates. Early in the PSA era (1987–1990), men in the Seattle area were screened and treated more intensively for prostate cancer than men in Connecticut. Objective  We previously reported more intensive screening and treatment early in the PSA era did not lower prostate cancer mortality through 11 years and now extend follow-up to 15 years. Design  Natural experiment comparing two fixed population-based cohorts. Subjects  Male Medicare beneficiaries ages 65–79 from the Seattle (N = 94,900) and Connecticut (N = 120,621) SEER areas, followed from 1987–2001. Measurements  Rates of prostate cancer screening; treatment with radical prostatectomy, external beam radiotherapy, and androgen deprivation therapy; and prostate cancer-specific mortality. Main Results  The 15-year cumulative incidences of radical prostatectomy and radiotherapy through 2001 were 2.84% and 6.02%, respectively, for Seattle cohort members, compared to 0.56% and 5.07% for Connecticut cohort members (odds ratio 5.20, 95% confidence interval 3.22 to 8.42 for surgery and odds ratio 1.24, 95% confidence interval 0.98 to 1.58 for radiation). The cumulative incidence of androgen deprivation therapy from 1991–2001 was 4.78% for Seattle compared to 6.13% for Connecticut (odds ratio 0.77, 95% confidence interval 0.67 to 0.87). The adjusted rate ratio of prostate cancer mortality through 2001 was 1.02 (95% C.I. 0.96 to 1.09) in Seattle versus Connecticut. Conclusion  Among men aged 65 or older, more intensive prostate cancer screening early in the PSA era and more intensive treatment particularly with radical prostatectomy over 15 years of follow-up were not associated with lower prostate cancer-specific mortality.     

Guillain-Barré syndrome after influenza vaccination in adults: a population-based study

BACKGROUND: Whether influenza vaccination is associated with Guillain-Barré syndrome (GBS) remains uncertain. METHODS: We conducted 2 studies using population-based health care data from the province of Ontario, Canada. In the first study, we used the self-matched case-series method to explore the temporal association between probable influenza vaccination (adults vaccinated during October and November) and subsequent hospitalization because of GBS. In the second study, we used time-series analysis to determine whether the institution of a universal influenza immunization program in October 2000 was associated with a subsequent increase in hospital admissions because of GBS at the population level. RESULTS: From April 1, 1992, to March 31, 2004, we identified 1601 incident hospital admissions because of GBS in Ontario. In 269 patients, GBS was diagnosed within 43 weeks of vaccination against influenza. The estimated relative incidence of GBS during the primary risk interval (weeks 2 through 7) compared with the control interval (weeks 20 through 43) was 1.45 (95% confidence interval, 1.05-1.99; P = .02). This association persisted in several sensitivity analyses using risk and control intervals of different durations. However, a separate time-series analysis demonstrated no evidence of seasonality and revealed no statistically significant increase in hospital admissions because of GBS after the introduction of the universal influenza immunization program. CONCLUSION: Influenza vaccination is associated with a small but significantly increased risk for hospitalization because of GBS.     

Two-stage methods for the analysis of pooled data.

Epidemiologic studies of disease often produce inconclusive or contradictory results due to small sample sizes or regional variations in the disease incidence or the exposures. To clarify these issues, researchers occasionally pool and reanalyse original data from several large studies. In this paper we explore the use of a two-stage random-effects model for analysing pooled case-control studies and undertake a thorough examination of bias in the pooled estimator under various conditions. The two-stage model analyses each study using the model appropriate to the design with study-specific confounders, and combines the individual study-specific adjusted log-odds ratios using a linear mixed-effects model; it is computationally simple and can incorporate study-level covariates and random effects. Simulations indicate that when the individual studies are large, two-stage methods produce nearly unbiased exposure estimates and standard errors of the exposure estimates from a generalized linear mixed model. By contrast, joint fixed-effects logistic regression produces attenuated exposure estimates and underestimates the standard error when heterogeneity is present. While bias in the pooled regression coefficient increases with interstudy heterogeneity for both models, it is much smaller using the two-stage model. In pooled analyses, where covariates may not be uniformly defined and coded across studies, and occasionally not measured in all studies, a joint model is often not feasible. The two-stage method is shown to be a simple, valid and practical method for the analysis of pooled binary data. The results are applied to a study of reproductive history and cutaneous melanoma risk in women using data from ten large case-control studies.     

Impact of clinical urgency,physician supply and procedural capacity on regional variations in wait times for coronary angiography

Background  

Despite universal health care, there continues to be regional access disparities to coronary angiography in Canada. Our objective was to evaluate the extent to which demand-side factors such as clinical urgency/need, and supply-side factors, as reflected by differences in physician and procedural supply account for these inequalities.