INTRODUCTION Despite poor agreement, neighbourhood income is used as a proxy for household income, due to a lack of data availability. We quantified misclassification between household and neighbourhood income and demonstrate quantitative bias analysis (QBA) in scenarios where only neighbourhood income is available in assessing income inequalities on colorectal cancer mortality. METHODS This was a retrospective study of adults with colorectal cancer diagnosed 2006-14 from Statistics Canada's Canadian Census Health and Environment Cohort. Neighbourhood income quintiles from Statistics Canada were used. Census household income quintiles were used to determine bias parameters and confirm results of the QBA. We calculated positive and negative predictive values using multinomial models, adjusting for age, sex and rural residence. Probabilistic QBA was conducted to explore the implication of exposure misclassification when estimating the effect of income on 5-year mortality. RESULTS We found poor agreement between neighbourhood and household income: positive predictive values ranged from 21% to 37%. The bias-adjusted risk of neighbourhood income on 5-year mortality was similar to the risk of mortality by household income. The bias-adjusted relative risk of the lowest income quintile compared with the highest was 1.42 [95% simulation interval (SI) 1.32-1.53] compared with 1.46 [95% confidence interval (CI) 1.39-1.54] for household income and 1.18 (95% CI 1.12-1.24) for neighbourhood income. CONCLUSION QBA can be used to estimate adjusted effects of neighbourhood income on mortality which represent household income. The predictive values from our study can be applied to similar cohorts with only neighbourhood income to estimate the effects of household income on cancer mortality.

中文翻译：

---

在结直肠癌患者队列中按社区对家庭收入进行暴露错误分类的概率偏倚分析。


引言 尽管协议不佳，但由于缺乏数据可用性，社区收入被用作家庭收入的代表。我们量化了家庭和社区收入之间的错误分类，并在评估结直肠癌死亡率收入不平等的情况下展示了定量偏倚分析 （QBA）。方法 这是一项回顾性研究，对 2006-14 年诊断为加拿大统计局加拿大人口普查健康和环境队列的成年结直肠癌患者进行研究。使用了加拿大统计局的邻里收入五分位数。人口普查家庭收入五分位数用于确定偏倚参数并确认 QBA 的结果。我们使用多项模型计算阳性和阴性预测值，并调整年龄、性别和农村居住地。进行概率 QBA 以探讨在估计收入对 5 年死亡率的影响时暴露错误分类的影响。结果 我们发现邻里收入和家庭收入之间的一致性差：阳性预测值从 21% 到 37% 不等。邻里收入对 5 年死亡率的偏倚调整风险与家庭收入的死亡风险相似。与最高收入五分位数相比，最低收入五分位数的偏倚调整相对风险为 1.42 [95% 模拟区间 （SI） 1.32-1.53]，而家庭收入为 1.46 [95% 置信区间 （CI） 1.39-1.54]，邻里收入为 1.18 （95% CI 1.12-1.24）。结论 QBA 可用于估计社区收入对代表家庭收入的死亡率的调整后影响。 我们研究的预测值可以应用于只有社区收入的类似队列，以估计家庭收入对癌症死亡率的影响。