(198) Diagnostic Suspicion Bias in Studies of Thiazide and Non-Melanoma Skin Cancer – A Cohort Study

Saturday, August 27, 2022

8:00 AM – 6:00 PM CEST

Location: Hall C

Publication Number: 872

Presenting Author(s)

Anat Fisher, MD, PhD
– Rsearch Associate, University of British Columbia, University of British Columbia, Vancouver, British Columbia, Canada

Co-Author(s)

Jason D. Kim
– University of British Columbia, Canada
Colin Dormuth
– University of British Columbia, Victoria, British Columbia, Canada

Background: Regulators have issued safety warnings for an association between thiazide use and non-melanoma skin cancer (NMSC)following observational studies that found an association between thiazide use and non-melanoma skin cancer (NMSC). We hypothesized that diagnostic suspicion bias could impact the significance or magnitude of this association.

Objectives: To estimate the impact of diagnostic suspicion bias in studies of thiazide and NMSC.

Methods: A cohort study of users of thiazides (exposed), angiotensin-converting enzyme inhibitors or angiotensin receptor blockers (controls), using administrative data from British Columbia (2000–2019). Users entered the cohort at the first-ever diagnosis of NMSC by a family practitioner (suspected NMSC) and were followed until NMSC diagnosis confirmation based on a validated combination of diagnoses and procedures. Exposed and control groups were frequency-matched (1:3) on year, age, sex, geographic area, income, and use of other photosensitizing medications. We estimated the hazard ratios using Cox regression, adjusting for the matching factors, comorbidities, and prior use of other medications and health services. A hazard ratio of less than 1 (exposed compared with controls) of a confirmed NMSC diagnosis supported an increased probability of a false-positive diagnosis among thiazide users with suspected NMSC, i.e., the presence of diagnostic suspicion bias.

Results: In the frequency-matched cohort, 578 of 3138 exposed individuals (18.4%) and 2010 of the 9477 controls (21.2%) had a confirmed NMSC diagnosis during 1 year of follow-up. The estimated crude and adjusted hazard ratios for the frequency-matched cohort were 0.96 (95% confidence interval 0.86, 1.07) and 0.96 (0.85, 1.09). In an analysis stratified by year of cohort entry (before 2017 when the large observational studies on the thiazide-NMSC association were published, or 2017 and later) the hazard ratio moved away from the null after 2017.

Conclusions: Our method of detecting and assessing the impact of diagnostic suspicion bias is generalizable to other populations and conditions. While our estimates of the bias did not reach the level of statistical significance, the direction of the estimates and the trends of findings after 2017 may indicate the presence of such a bias.