(B-198) Initial-treatment Estimates as Negative Controls for Baseline Confounding Under Low Treatment Persistence: a Real-World Application to Opioids and Diverticulitis

Background: Negative control outcomes (i.e., outcomes the exposure is known not to affect ) are commonly used in observational studies to evaluate whether the observed exposure-outcome association may result from confounding by unmeasured variables . Identifying a suitable negative control outcome is challenging as associations between confounders, the exposure, and both the primary and negative control outcome should be similar. Initial-treatment (IT) outcomes may satisfy these conditions when as-treated (AT) effects are of interest under low persistence and several other conditions.

Objectives: To describe the use of IT analyses as a negative control for AT analyses in a real-world study comparing hospitalized diverticulitis risk in new users and non-users of opioids.

Methods: Whether researchers are estimating IT or AT effects, the treatment and comparator groups of interest are on average identical at baseline in both analyses. If the proportion of patients persisting on treatment is very low, the treatment takes time to cause harm, treatment effects wear off soon after cessation, and few in the comparator group later initiate the treatment of interest, the IT effect should be close to the null . In such scenarios, the IT effect fulfills the conditions for a negative control for baseline confounding. We demonstrated this using data from 5 Canadian provinces, the United Kingdom (UK), and the United States (US) to examine the association between new use of opioids and hospitalization for diverticulitis. We estimated 2-year AT and IT risk ratios (RRs) comparing opioid new use to non-use following an elective surgical procedure after odds weighting to balance covariates and pooling estimates using random effects meta-analysis.

Results: The analysis included 4,594,707 patients from Canada, 787,200 from the UK’s Clinical Practice Research Datalink, and 17,702,503 from the US Merative MarketScan database. Average AT follow-up for opioid new users was 42 days, with a heavily left-skewed distribution indicating many patients stopped treatment after receiving initial prescriptions. The pooled 730-day AT RR estimate was 2.22 (95% CI: 1.08, 4.54). The corresponding IT estimate was 0.95 (95% CI: 0.88, 1.04).

Conclusions: By employing IT analysis as a negative control for the AT effect of opioids on diverticulitis, our findings suggest baseline confounding alone is unlikely to explain the association between sustained new opioid use and hospitalized diverticulitis. When persistence with therapy is low, IT effects may serve as negative control outcomes with respect to the influence of baseline confounding on AT treatment effects. IT effects cannot, however, be used to rule out selection bias.