0042 Correction for reporting bias: the importance of stratum specific estimates. (23rd June 2014)

• Record Type: Journal Article
• Authors: Cherry, Nicola; Beach, Jeremy; Burstyn, Igor; Sentilselvan, Ambikaipakan
• Is Part Of: Occupational and environmental medicine. Volume 71(2014)Supplement 1
• Page Start: A3
• Page End: A3
• Publication Date: 2014-06-23
• Subjects: Medicine, Industrial -- Periodicals; Environmental health -- Periodicals; 616.980305
• DOI: 10.1136/oemed-2014-102362.9
• Languages: English
• ISSNs: 1351-0711

Abstract

Abstract : Objectives: To examine whether reporting bias was responsible for increasing rates of self reported heart disease (SRHD) with exposure to phenoxy herbicides in a cohort of elderly grain farmers in Alberta. Method: We estimated exposure to chemical ingredients from named pesticides reported at interview after a lifetime of farming. Phenoxy exposure was grouped by tertiles as none, 1–22 yrs, 23–34 yrs and 35 yrs or greater. Six years after interview consent was sought from surviving farmers to link questionnaire data to provincial physician billing records. Sensitivity (Se) and specificity (Sp) were estimated for SRHD, overall and within exposure stratum, and adjusted odds ratios (ORs) calculated. Results: Among the cohort of 2426, 373 had SRHD: ORs, adjusted for confounding, were estimated for phenoxy tertile (compared with no exposure) as 1.32 (95% CI 0.79–2.23), 1.67 (0.99–2.85), 2.03 (1.20–3.45). For the 1371 farmers consenting to record linkage, ORs before adjustment or correction were 1.25, 1.43, 2.69. Comparing SRHD to physician billing diagnoses of heart disease gave an overall Se 0.49 and Sp 0.98. ORs were increased to 1.37, 1.65, 4.01 when corrected by these Se and Sp estimates. Differences were seen in stratum specific Se but not Sp, with significantly lower Se (0.29: 95% CI 0.15–0.48) in the group with no exposure. Applying stratum specific correction reduced ORs in all three categories (0.55, 0.82, 1.29). Conclusions: The correction approach used, which demonstrated the importance of stratum specific estimates, assumed no error in the validation data. Sensitivity analyses to explore this limitation will also be presented.