Successful Implementation of Inverse-probability-of-screening Weights to Address Differential Outcome Screening: Effects of Statin Persistence on Breast Cancer Risk in Denmark

Background: Statins have shown antineoplastic activity in pre-clinical studies, and several studies with long-term follow-up have shown a small risk-reducing impact of statins on breast cancer (BrCa) risk. However, these studies may be biased by healthy adherer effects and did not address how differential use of screening mammography during follow-up might affect BrCa incidence.

Objectives: We sought to evaluate whether initial statin persistence was associated with 1) screening mammography and 2) risk of incident BrCa accounting for any differential screening behavior in a contemporary Danish statin initiator cohort.

Methods: We identified female statin initiators aged 50-69 in Denmark between 2001-2018 with no history of BrCa, and classified women according to their statin persistence in the first 6 months after initiation. We estimated age- and calendar time-standardized mammography coverage (i.e., proportion screened) by persistence. We then estimated BrCa risk controlling for baseline confounding using inverse-probability-of-treatment weights (IPTW) and using inverse-probability-of-screening weights (IPSW) to account for differential detection. For the IPSW, we estimated a screening propensity score in each interval based on screening predictors in the prior interval; these included persistence, prior screening, family history of breast cancer, other screening tests (e.g., bone mineral density scans), comorbidities, and comedications. The inverse of this score served as the denominator of the IPSW, which was stabilized by the marginal screening frequency. The cumulative product of all prior interval IPSW was used to account for longitudinal screening behavior.

Results: A total of 267,390 women met study inclusion criteria, and 81.1% of the cohort was persistent. Persistent women had 5.7 percentage points higher screening coverage over follow-up (52.2% vs. 46.5%). Statin persistence and prior screening were associated with higher probability of subsequent screening, while hospitalizations and smoking- or alcohol-related diagnoses were negatively associated. IPSW largely removed these differences. In crude analysis, the 15-year risk of BrCa was 0.05; persistent women had a 15-year BrCa risk difference (RD) of 0.002 (95% confidence interval [CI]: -0.001, 0.005). The RD was -0.003 (95% CI: -0.014, 0.008) after implementing IPTW/IPSW.

Conclusions: Statin persistent women had higher screening coverage, leading to a slightly higher observed BrCa risk in crude analysis; accounting for differential screening produced a negative observed RD, in line with the hypothesized healthy adherer bias, though all estimates remained nonsignificant. We encourage investigators to implement IPSW whenever differential outcome surveillance is suspected.