Our study found increased mortality as educational years decreased, particularly high differentials for respiratory diseases, infectious diseases, diabetes and homicide and lower for cancer and suicide. Arabs had similar educational differences as Jews and Others, and lower risk of mortality for suicide, cancer in females, infectious diseases and dementia in males, and significantly higher risk for diabetes, homicide, heart and cerebrovascular diseases, and respiratory diseases in males, after controlling for educational status.
How does education affect morbidity and mortality? Phelan et al. [25] suggest multiple mechanisms for the effect of socio-economic status on health, such as influencing disease outcomes, through multiple risk factors and affecting access to resources that can be used to minimize risk factors or treat existing disease. Let us look at whether our findings confirm education as performing these mechanisms showing it to be a correct social determinant of health in Israel.
Firstly, we saw (Additional file 1: Table S5) that the more educated in general had a healthier lifestyle and lower risk factors for disease, probably due to better knowledge and the economic ability to act accordingly. They smoked less, had lower BMI, were on weight reducing diets more often, ate more vegetables, drank less sweet drinks and did more physical exercise (although there were some exceptions to these trend for the lowest educational group). These differences in behavior appear to be reflected in the high HRs we found for the more ‘preventable’ diseases of diabetes, heart disease and cerebrovascular disease, and similar to the high proportion of premature deaths for these diseases attributable to socio-economic inequalities found by Lewer et al. [26], and strongest educational gradient found by Steenland et al. [8].
Educational level has economic consequences, with higher education giving better employment opportunities, income, housing, and access to healthcare, shown by the increasing rates of supplementary health insurance with education. Although Israel has universal health insurance that covers most health care, we see that the lower educated reported more often doing without prescription medicine or having medical treatments, which could contribute to worse outcomes for many diseases. Crowded housing may lead also to more infectious diseases, for which we found a high HR.
As noted by Hummer et al. [9], we saw a strong influence on causes linked with social and behavioural risk factors, such as homicide and respiratory diseases, the causes with the highest HR in males.
A lesser effect of education was found on causes less amenable to control and prevention, such as cancer. In addition, some screening tests for cancer, such as mammograms and testing for faecal occult blood are covered by the Israeli health ‘basket’ of services, and recent health policy encourages their implementation. Indeed, Additional file 1: Table S5 showed little difference in frequency of mammograms between the educational groups, and the occult blood test was more frequent at lowest educational level, although this could be because the better educated used colonoscopies instead, more frequent among them. The PAP smear test, however, does show the educational gradient, but may be less important for the lower educational groups, in view of their lifestyle. Breast cancer, in particular, has been reported higher in the higher educated, although Trewin et al. have shown the mortality trend to have begun reversing in Norway [27]. Nevertheless, as the leading cancer in women, this may contribute to the lack of significant difference found for cancer mortality among lowest educated women. The relatively low HR for cancer is similar to that found in other studies, such as Yang et al. [7].
There may also be an effect of greater work stress on those with higher education which may offset some of their advantage, as reported in a meta-analysis by Yang et al. that some cancers, such as colorectal and lung, were associated with work stress [28]. We also note, for example, that the higher educated ate 2 or more portions of wholegrain food less frequently than the lesser educated.
Suicide is one cause that seems to be lesser affected by educational status, and in females, the effect is non-significant.
We thus see that our study supports education as a social determinant of health. The value of education as a socio-economic indicator is also since it is usually attained early in life and more likely to be a cause of better health than income or occupation which may have a two-way relationship with heath status [3]. In Israel, a disadvantage of education as a measure is that some of the population may have had limited opportunity for formal education due to the effect of the Holocaust, World War II or immigration from other countries, but nevertheless achieved high socio-economic status through informal education, or despite this lack of education. In addition, it is difficult to quantify the effect of religious studies, such as in Yeshivot, which do not enter the regular educational attainment measures. Social capital can also offset low formal educational achievements, which can lead to better health outcomes, as reported by Chernikovsy et al. [29].
Arabs compared to Jews and Others
Life expectancy at all ages has been higher for Jews than Arabs throughout the years and our age-adjusted model found significantly higher all-cause mortality and for mortality from most causes for Arabs. However, we found that after controlling for educational status too, there was only a slightly higher all-cause mortality risk for Arab males (HR = 1.07, 95% CI 1.05–1.09), and a non-significant comparative risk for females while for the younger cohort, aged 25–54 in 2000 (Additional file 1: Table S3), the adjusted mortality risk was lower for Arab females than Jews and Others (HR = 0.88, 95% CI 0.85–0.91) and insignificant for males. This would suggest that much of the difference in overall mortality between Jews and Others and Arabs can be attributed to educational differences. The effect of education was particularly strong in mitigating differences for respiratory diseases, homicide and diabetes and even more causes in females.
When we looked at the cause specific risks of Arabs compared to Jews and Others, we see differences even after adjusting for education, such as with much higher mortality from diabetes, particularly in females, heart disease, and homicide and respiratory disease in males—the latter reflecting the high smoking rate among Arab males [30]. These high risk causes are partially offset by significantly lower risks for suicide, dementia in males, infectious diseases, and cancer in females. The lower risk for cancer in Arab females reflects their lower risk for breast cancer [30], the leading female cancer. However, since Arab males have a higher incidence of lung cancer [30], the leading cancer for male mortality, probably also due to their high smoking levels, we found their comparative risk for cancer mortality compared to Jews and Others was insignificant.
Therefore, we would suggest that increasing educational level can be an effective intervention to improve Arab health, but this needs to be supplemented by ethnically appropriate programs targeting high smoking in males and high diabetes rates.
Comparison with previous Israeli studies—changes over time
Jaffe et al. [15] reported an increase in educational disparities between the lowest and highest educated groups, between census cohorts of Jews and Others aged 25–64 from 1983 and 1995, for all-cause and particularly for CVD mortality in females. In a similar group of Jews and Others (Fig. 2, Additional file 1: Table S3), we did not find a further increase in our 17 year follow-up from 2000, but rather our HRs were lower than the ORs of Jaffe et al., 1.89 (95%CI 1.86–1.93) for all-cause mortality for males and 1.58 (95%CI 1.86–1.93) for females, compared to 2.09 (95%CI 1.91–2.28) 2.02 (95%CI 1.81–2.25), respectively, found by Jaffe et al. Although our HRs for CVD were higher for females than males, 2.49 (95%CI 1.30–2.69) for heart disease and 2.30 (95%CI 1.30–2.69) for cerebrovascular disease they were less than half what Jaffe et al. found for CVD mortality, 5.14 (95%CI 3.48–7.60).
The narrowing of educational disparity particularly for CVD compared to Jaffe et al. is good news. We would suggest that the reason may be due to the steep decline in CVD as leading causes of death [19] over the period of our study, which was just beginning at the end of Jaffe et al.’s follow-up of causes of death in 2000. This was fueled by increasing awareness and treatment of risk factors, such as hypertension and hyperlipidemia, and better interventions and treatment for CVD, which became somewhat more accessible and available to those of lower socio-economic and educational levels than in 1995–2000. We hope this trend will continue to further reduce the gap in CVD mortality between the higher educated and the lower.
Strengths and limitations
The strength of this study is the large nationwide cohort followed up for 17 years, allowing analysis of educational differences in mortality for a range of causes of death with significant results.
The first limitation of our study was the missing data for education, particularly at older ages. Missing educational data has been reported as a problem in other studies too [3], and can be addressed in two ways. The persons with missing data are sometimes assigned to the lowest educational category, which could be supported in our study by their cause of death profile (Additional file 1: Table S1), similar to that of the lowest educated group. This method was adopted by the CBS in their publication [22], which includes results for mortality in our cohort, similar to those we found. In this paper, we chose the other, perhaps more exact way of handling them, by excluding them from the study, which might lead to an under-estimation of the lower educated. However, when we checked a younger cohort which had a smaller number missing education, as a sensitivity analysis, we found similar results.
A second limitation was not having individual level data on risk factors and economic and health indicators, such as smoking, hypertension, hyperlipidemia, income and occupation, which could have been mediators for the educational effects we found, as found by Steenland et al. [8].
Although we did multiple comparisons, due to our large study cohort, most of our results were highly significant (p-value < 0.0001), and would remain significant after correction for multiple comparisons.
Another limitation is that the proportional hazard assumption was not satisfied for all covariates in all models. However, the changes over time indicated by the estimated parameters for the time dependent variables that were significant did not lead to any major changes in our findings.
Health policy implications and recommendations
We found higher mortality in those with lower educational status. This needs to be addressed in two ways. Firstly, increasing the educational level of the population should help improve population health. Secondly, health-promoting interventions need to be provided in particular to the lesser educated, encouraging smoking cessation, eating a healthy diet and exercising regularly. The educational level in Israel has been increasing: according to the Statistical Abstract of the CBS [31] those with 13 years or more of education increased from 45 to 55% for Jews and Others between 2005 and 2017 and from 19 to 26% for Arabs. It is to be hoped that improved education will help close the health gap in Israel, but needs to be encouraged in all population groups to reduce future inequalities.
