Pharmacoepidemiology for nephrologists (part 2): potential biases and how to overcome them

Abstract

Observational pharmacoepidemiological studies using routinely collected healthcare data are increasingly being used in the field of nephrology to answer questions on the effectiveness and safety of medications. This review discusses a number of biases that may arise in such studies and proposes solutions to minimize them during the design or statistical analysis phase. We first describe designs to handle confounding by indication (e.g. active comparator design) and methods to investigate the influence of unmeasured confounding, such as the E-value, the use of negative control outcomes and control cohorts. We next discuss prevalent user and immortal time biases in pharmacoepidemiology research and how these can be prevented by focussing on incident users and applying either landmarking, using a time-varying exposure, or the cloning, censoring and weighting method. Lastly, we briefly discuss the common issues with missing data and misclassification bias. When these biases are properly accounted for, pharmacoepidemiological observational studies can provide valuable information for clinical practice.

Keywords: bias; causal inference; confounding; epidemiologic methods; observational studies; pharmacoepidemiology.

FIGURE 1

(A) Confounding by indication arises when prognostic factors for the outcome also influence the decision to start treatment. Some confounders may be measured, which can be adjusted for in the analysis, whereas others are unmeasured, leading to residual confounding. (B) When unintended outcomes are studied, less confounding by indication will be present. The indications for ACEi treatment likely do not increase the risk for the outcome angioedema. (C) An ideal active comparator has similar indications as the medication under study, thereby decreasing confounding by indication. Ideally the active comparator should have no influence on the outcome. (D) Negative control outcomes need to have similar measured and unmeasured confounders as the treatment–outcome relationship under study. Furthermore, treatment should not have an influence on the negative control outcome.

FIGURE 2

Graphical visualization of (A) prevalent-user bias and (B) immortal time bias when setting up the start of follow-up in a study. For prevalent-user bias, the start of follow-up occurs after treatment initiation, whereas for immortal time bias, the start of follow-up occurs before treatment initiation. These biases can be prevented by aligning the start of follow-up with the start of exposure.

FIGURE 3

Design of a landmark analysis to prevent immortal time bias. In the landmark analysis, follow-up starts at a chosen time period after a certain event, in this example at 6 months. Hence all individuals that died before Month 6 are excluded from the analysis (Individual 4). Individuals are then classified according to exposure status in the first 6 months. Individuals 3, 5 and 6 are therefore considered treated, whereas Individuals 1 and 2 are considered untreated.

FIGURE 4

Analysis using a time-varying exposure to prevent immortal time bias. In a time-varying design, treatment status is allowed to change from unexposed to exposed at the moment of treatment initiation. This method allows the start of follow-up directly after the event has occurred and also does not exclude individuals. For example, Individual 1 is considered unexposed for the first 7 months of follow-up, but after 7 months will contribute to the exposed group. In the setting of time-varying exposures, time-varying confounding will be present too, which sometimes requires more advanced methods, such as marginal structural models, to obtain unbiased effect estimates.

FIGURE 5

Design of a study using treatment strategies with a grace period based on cloning, censoring and weighting. Another method is comparing treatment strategies that include a grace period. Each individual is duplicated and assigned to one of two treatment strategies. In this example, Clones 1a–6a follow the strategy ‘initiate ACEi within 6 months’, whereas Clones 1b–6b follow the strategy ‘do not initiate ACEi within 6 months’. Note that Copies 1a and 1b represent the same individual. Since Copy 1a is assigned to initiating within 6 months, he is censored after Month 6, as he did not initiate treatment. The censoring is likely to be informative and inverse probability weighting is required to adjust for this.