An unadjusted comparison can point in one direction while an adjusted analysis points in the other. That happened for methamphetamine and incident atrial fibrillation in a large California cohort. Understanding the change requires looking at who was compared, not treating the largest dataset as automatically immune to confounding.
What did the study report?
Lin and colleagues followed 23,561,884 adults receiving California hospital-based care during 2005–2015. The dataset identified 98,271 with methamphetamine-use coding. Existing atrial fibrillation at the first encounter was excluded. After adjustment, the methamphetamine association with new atrial fibrillation was a hazard ratio of 1.86 (95% CI 1.81–1.92). Lin et al., 2022.
The authors found age most responsible for the reversal from a lower unadjusted association to a higher adjusted association. Exposure began at the first coded substance-use encounter. Dose-response could not be established from the codes. Documentation errors, incomplete ascertainment and residual confounding remained limitations. The authors expressly cautioned that dataset size does not eliminate bias. Methods, results and discussion.
How can a reversal occur?
Imagine comparing two groups with markedly different age distributions. The crude comparison mixes the exposure question with the age difference. A younger exposed group can have fewer observed events overall even if, within comparable ages, exposure is associated with more events. Adjustment attempts to separate those influences using the model and measured variables.
This example explains the statistical possibility; it is not a reconstruction of unpublished individual patient data. A reversal does not make either number fraudulent. It makes the composition of the groups essential to understanding what each number answers.
Does adjustment recover the true causal effect?
Not automatically. Adjustment depends on which characteristics were measured, how accurately they were measured and how the model represents them. Some variables may lie on a pathway between exposure and disease, while others precede both. Controlling for them can therefore change the causal question as well as the estimate.
A careful interpretation names the adjusted association and its population. It does not translate “after adjustment” into “all alternative explanations excluded.” Nor does it treat the change in direction as evidence that a specific biological mechanism was measured.
What does a hazard ratio of 1.86 mean?
It is a relative time-to-event estimate under the study’s model. It is not an 86% absolute probability of developing atrial fibrillation and cannot be applied as a percentage of all methamphetamine exposures. Absolute risk additionally depends on baseline risk, follow-up and the population under consideration.
The outcome is a newly recorded diagnosis. The date that a disorder is first coded need not be the precise biological onset. Similarly, a coded exposure history does not provide a blood concentration, administered amount or proof of intoxication at the moment an arrhythmia occurred.
How should this affect an individual opinion?
The study supplies population-level evidence supporting investigation of an association. Individual attribution still requires the actual rhythm documentation, prior cardiac history, exposure chronology and competing explanations. A group hazard ratio cannot substitute for that clinical reconstruction.
This historical observational evidence was published in 2022. Its methodological lesson is unusually concrete: a large crude comparison can mask an association when the exposed and comparison populations differ in age. The adjusted finding is informative, while its causal limits remain.
Source
Lin AL, Nah G, Tang JJ, Vittinghoff E, Dewland TA, Marcus GM. Cannabis, cocaine, methamphetamine, and opiates increase the risk of incident atrial fibrillation. European Heart Journal. 2022;43:4933–4942. doi:10.1093/eurheartj/ehac558.