The social patterning of measurement errors in self-reports: Impact on socioeconomic inequalities estimates

Abstract

Introduction Reduction of unfair differences in health between socioeconomic groups and countries constitutes an important public health challenge in the 21st century. To monitor progress on this goal, health inequalities are most frequently estimated based on self-reported data from population surveys. However, it has been shown that self-reported data on cardiovascular disease risk factors is prompt to reporting error. If errors occur more often in specific socioeconomic groups (due to under-diagnosis or lower literacy) they are likely to seriously bias health inequality estimates. This study aims at comparing measurement errors between socioeconomic categories in self-reported hypertension, and their consequences on health inequality estimates. Methods We used data from the Portuguese National Health Examination Survey (INSEF), a cross-sectional nationwide study conducted in 2015 on a probabilistic sample (n = 4911) of community-dwelling individuals aged between 25 and 74-years-old. INSEF combines measured biochemical parameters and blood pressure with self-reported data. Self-reported hypertension was defined based on the question: “Do you have any of the following diseases or conditions: hypertension? (Yes/No)”. Examination-based hypertension was defined as having systolic blood pressure of at least 140 mmHg, or diastolic blood pressure of at least 90 mmHg, or using prescribed antihypertensive medication. Participants’ socioeconomic status was measured through the education level (no education/1st cycle of basic education, 2nd cycle of basic education, 3rd cycle of basic education, secondary education and higher education). Inequalities in hypertension between the highest and lowest socioeconomic status groups were measured using relative indexes of inequality (RII) and respective confidence intervals (95% CI), estimated by Poisson regression. Estimates of inequalities were stratified by age and sex, using four population groups (male 25–49-years-old, female 25–49-years-old, male 50–75-years-old, female 50–75-years-old). Results Hypertension was reported by 25.7% [95% CI: 24.0 to 27.4] of participants, while 35.9% [95% CI: 34.2 to 37.5] were considered to have hypertension according to examination-based data. The difference between examination-based and self-reported prevalence was 12.7pp for those with no education/1st cycle of basic education and 4.6pp for those with higher education. Similar educational gradients were observed for both self-reported (RII = 1.87; 95% CI: 1.45 to 2.42) and examination-base (RII = 1.91; 95% CI: 1.60 to 2.28) hypertension, with lowest prevalence of disease among the highly educated. Age- and sex-specific results showed considerable discrepancies in inequality indicators between self-reported and examination-based data. Namely, differences in estimated gradients were more pronounced among 25–49-years-old males, with RII = 0.67 (95% CI: 0.29 to 1.54) for self-reported and RII = 1.90 (95% CI: 1.22 to 2.96) for examination-based hypertension. In 25–49-years-old females inequalities in self-reported hypertension were not statistically significant (RII = 3.18; 95% CI: 0.94 to 10.73), while females with the lowest education were 4.35 (95% CI: 2.60 to 7.27) times more likely to have examination-based hypertension then compared to the most educated. In 50–75 age group educational inequalities in self-reported hypertension were larger than in examination-based for both, male (RII = 1.82; 95% CI: 1.25 to 2.69 vs. RII = 1.40; 95% CI: 1.04 to 1.89) and female (RII = 1.77; 95% CI 1.30 to 2.41 vs. RII = 1.58; 95% CI: 1.22 to 2.04). Conclusions Our results illustrated the significant effect of measurement error in self-reported hypertension on estimates of socioeconomic inequalities. Use of self-reported data led to underestimation of educational inequalities among young and middle-aged individuals and overestimation in older age groups. Inequality indicators derived from self-report should be interpreted with caution.