Using inverse probability-weighted estimators in comparative effectiveness analyses with observational databases.

Duke Clinical Research Institute, and Departments of Medicine, Duke University School of Medicine, Durham, North Carolina, USA.
Medical Care (Impact Factor: 2.94). 11/2007; 45(10 Supl 2):S103-7. DOI: 10.1097/MLR.0b013e31806518ac
Source: PubMed

ABSTRACT Inverse probability-weighted estimation is a powerful tool for use with observational data. In this article, we describe how this propensity score-based method can be used to compare the effectiveness of 2 or more treatments. First, we discuss the inherent problems in using observational data to assess comparative effectiveness. Next, we provide a conceptual explanation of inverse probability-weighted estimation and point readers to sources that address the method in more formal, technical terms. Finally, we offer detailed guidance about how to implement the estimators in comparative effectiveness analyses.

  • [Show abstract] [Hide abstract]
    ABSTRACT: Current evidence is mixed regarding the association between antihypertensive prescriptions and cancer mortality. We evaluated this association in a large Chinese hypertensive population.We followed for five years all patients who were prescribed their first-ever antihypertensive agents between 2001 and 2005 in a public healthcare sector of Hong Kong. The association between antihypertensive drug class and cancer mortality was evaluated by Cox proportional hazard models with propensity score matching. Age, gender, socioeconomic status, service settings, district of residence, proportion of days covered reflecting medication adherence, and the number of comorbidities were adjusted.From 217,910 eligible patients, 9500 (4.4%) died from cancer within five years after their first-ever antihypertensive prescription. Most cancer deaths occurred in the digestive (38.9%) and respiratory system (30.4%); the breast (6.2%); and the lympho-hematopoietic tissues (5.3%). The proportion of patients who died from cancer was the highest in the calcium channel blocker (CCB) group (6.5%), followed by thiazide diuretics (4.4%), angiotensin converting enzyme inhibitors (4.2%) and β-blockers (2.6%). When compared with β-blockers, patients prescribed CCBs (Adjusted Hazard Ratio [AHR] = 1.406, 95% C.I. 1.334–1.482, p < 0.001) were more likely to die from cancer. Thiazide users were also more likely to suffer from cancer deaths (AHR = 1.364, 95% C.I. 1.255–1.483, p < 0.001), but became insignificant in stratified analysis.The association between cancer mortality and use of CCB, and perhaps thaizide, may alert physicians to the need for more meticulous and comprehensive care of these patients in clinical practice. We recommend prospective studies to evaluate cause-and-effect relationships of these associations.
    International Journal of Cardiology 10/2014; · 6.18 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: 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.
    Health Services and Outcomes Research Methodology 06/2014; 14(1-2).
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: As providing health education, optimizing nutrition, and managing risk factors can be effective for ensuring a healthy outcome for women and her yet un-conceived baby, external influences play a significant role as well. Alcohol, smoking, caffeine use and other similar lifestyle factors, have now become an integral part of the daily life of most men and women, who use/misuse one or more of these harmful substances regularly despite knowledge of their detrimental effects. The adverse health outcomes of these voluntary and involuntary exposures are of even greater concern in women of child bearing age where the exposure has the potential of inflicting harm to two generations. This paper is examining the available literature for the possible effects of caffeine consumption, smoking, alcohol or exposure to chemicals may have on the maternal, newborn and child health (MNCH). A systematic review and meta-analysis of the evidence was conducted to ascertain the possible impact of preconception usage of caffeine, tobacco, alcohol and other illicit drugs; and exposure to environmental chemicals and radiant on MNCH outcomes. A comprehensive strategy was used to search electronic reference libraries, and both observational and clinical controlled trials were included. Cross-referencing and a separate search strategy for each preconception risk and intervention ensured wider study capture. Heavy maternal preconception caffeine intake of >300mg/d significantly increase the risk of a subsequent fetal loss by 31% (95% CI: 8-58%). On the other hand, preconception alcohol consumption leads to non-significant 30% increase in spontaneous abortion (RR 1.30; 95% CI: 0.85-1.97). Preconception counselling can lead to a significant decrease in the consumption of alcohol during the first trimester (OR 1.79; 95% CI: 1.08-2.97). Periconception smoking, on the other hand, was found to be associated with an almost 3 times increased risk of congenital heart defects (OR 2.80; 95% CI 1.76-4.47). While the review found limited evidence of preconception environmental exposure on maternal, newborn and child health outcomes, occupational exposure in female radiation workers before conception showed an increased impact in risk of early miscarriages. Identification of substance abuse and environmental history during preconception period provides an opportunity to assist women in reducing major health risks and identify key determinants of healthy pregnancy. Studies have shown that the aversion and prevention of exposure feasibility can play an important role in improving the health of women and their families, however, the results should be interpreted with great caution as there were few studies in each section. Therefore, there is a need for more rigorous studies to test the hypotheses.
    Reproductive Health 09/2014; 11 Suppl 3:S6. · 1.62 Impact Factor

Full-text (2 Sources)

Available from
May 22, 2014