Socioeconomic inequality in self-reported unmet need for oral health services in adults aged 50 years and over in China, Ghana, and India

The data source, WHO SAGE Wave 1 (2007–2010), is a longitudinal study of nationally representative samples of adults aged 50 years and over in China, Ghana, India, Mexico, Russia and South Africa. In this study we draw on the cross-sectional results of WHO SAGE Wave 1 national surveys implemented between 2007 and 2010 in China, Ghana and India. Data were obtained from self-reported responses to validated structured household and individual questionnaires.

WHO-SAGE Wave 1 used a stratified random sampling frame. Sampling weights were applied to adjust for age and sex distributions, urban and rural localities and non-response. In China the sample was stratified into eight nationally representative provinces. The Ghana sample was stratified by administrative region and urban or rural locality, giving eighteen nationally representative strata. In India states were selected in accordance with geographic location and level of development. The sample in India was stratified by state and urban or rural locality giving 12 nationally representative strata [41].

Data collection in Ghana and India was through face-to-face paper and pencil interviews. In China 50% of the data were collected by face-to-face computer assisted personal interviews and 50% by face-to-face paper and pencil interviews [41]. A standardised questionnaire was used in all countries. Questionnaires were translated into local languages, validated and then back-translated. Interviews were conducted by trained interviewers in culturally appropriate settings. WHO protocols were closely monitored throughout data collection [42].

WHO-SAGE country data sets and sampling weights, are in the public domain. Further details of WHO-SAGE are provided elsewhere [43].

WHO SAGE Wave 1 included 47,443 respondents aged 18 and over. This analysis covers samples drawn from SAGE respondents aged 50 and above in China (n = 15,050), Ghana (n = 5565) and India (n = 12,198). Response rates were 93, 80, and 68% in respectively [44]. Respondents less than 50 years of age in China (1636), Ghana (805) and India (4670) were excluded as were those who did not complete the WHO-SAGE individual questionnaire - China (n = 237), Ghana (463) and India (968).

Respondents who completed the surveys and were aged 50 years and over were: 13,177 in China, 4297 in Ghana and 6560 in India. The country samples were further conditioned on self-reported need for oral health services elicited from two key questions outlined in the section below. Only those respondents who expressed a need for oral health services were included giving 1607 in the China sample, 431 in the Ghana sample and 1316 in the India sample. A further 31 records were excluded due to missing data on key variables (China 16, Ghana 6 and India 9).

The Mexican (n = 5548) Russian (n = 4947) and South African (n = 4227) samples were not analysed here because of high percentages of missing data on the study variables (52 and 13% in Mexico and Russia respectively) and small sample sizes after conditioning for self-reported need for oral health (e.g. 81 in South Africa). See Fig. 1.

The dependent variable, unmet need for oral health services, was derived from self-reported answers to two survey questions: 1) “Have you had any problems with your mouth/teeth during the last 12 months?” and 2) “Have you received any medication or treatment from a dentist or any other oral health specialist during the last two weeks or 12 months?” Only respondents who answered “yes” to the first question were considered for the second question. A binary indicator variable (unmet need) was derived from answers to the second question - those who responded “yes” were classified as having no unmet need for oral health services in the prior 12 months and those who responded “no” were classified as having unmet need in the prior 12 months.

Two categorical independent variables – household wealth and education – were the socioeconomic measures used to quantify inequality in self-reported unmet need for oral health services.

The household wealth variable was based on information on ownership of household assets reported in the WHO-SAGE household questionnaire. This included ownership of durable goods (e.g. car, television, radio, and refrigerator), household characteristics (e.g. type of windows, roof and floor material), and access to basic services (e.g. source of water and electricity). Principal component analysis was used to generate weights for household assets [45]. Wealth quintiles were produced with quintile one representing the lowest wealth and quintile five the highest. In this study, the wealth quintiles were recoded with quintile one representing the highest wealth and quintile five the lowest.

The education variable was derived from two survey questions “Have you ever been to school?” and “What is your highest level of education completed?” The variable indicated the highest reported level of education and was categorized in four groups; university or higher, secondary, primary, and no schooling.

Covariates were: sex (male or female); age categorized as 50–59 years, 60–69 years, 70–79 years, and 80 years and over; area of residence (urban or rural); marital status (never married versus married or cohabiting versus widowed, divorced or separated); work status (never worked versus currently working versus currently not working), and self-rated health (good versus moderate versus bad).

Only records with complete data on all study variables were included in these analyses. Survey sampling weights were used in all analyses. Descriptions of the country samples are given by absolute numbers and weighted percentages. The prevalence of unmet need was estimated with 95% confidence intervals (CIs) to allow statistical comparison between countries.

The relative index of inequality (RII) was used to separately measure inequalities in unmet need for oral health services in terms of household wealth and education. The RII is a summary measure which quantifies relative socioeconomic inequalities in an outcome by taking into account the size of the population across ordered socioeconomic categories [46]. Here we have two RIIs – one for household wealth and one for education. The RII is defined as the ratio of the estimated prevalence of unmet need (for oral health services) between the poorest and wealthiest or between the least educated and the most educated. If, for example, the RII for wealth is 1.6 and statistically significant this means that the estimated prevalence of unmet need is 1.6 times higher among the poorest group compared with wealthiest group. If the RII is close to one (or not statistically significant) then we can say that there is no evidence of socioeconomic inequality. If the score is less than one and significant, for example 0.5, then in this example, the unmet need would be 50% lower in the poorer group.

One of the strengths of the RII is that it takes the size of the population across socioeconomic categories into account. In order to compute the RII it is necessary to arrange the sample into the socioeconomic categories. Each group is characterized by a “ridit” which corresponds to the average cumulative frequencies of each of the groups [46]. The RII is computed by regressing ridit scores on the outcome. This makes it possible to adjust for possible confounding by undertaking multivariable analyses. The adjusted models control for the possible confounding effects of sex, age, area of residence, marital status, work status and self-rated health in self-reported unmet need.

We did not conduct a pooled multi-country analysis with a “country” variable for a couple of reasons. First, a large number of countries would have been needed to estimate country effects reliably. Second, the country sample was defined in accordance with the way in which respondents answered questions about oral health services and we were concerned that differences in the distributions of responses (to the oral health questions) in each of the countries would have led to bias in a pooled analysis.

Coefficients were obtained by Poisson models using the logarithmic link function expressed as exp.(β). All statistical analyses were carried out using STATA 13 software (StataCorp, 2013).

The distribution of selected demographic and socioeconomic characteristics in the country samples is given in Table 1. The proportions of females to males were just above 50% in China and India and just under 50% in Ghana. Just over one-third of the adults were aged 50–59 years. Approximately 62% in China and Ghana, and 73% in India were rural residents. Around 20% of the adults belonged to the lowest wealth group in each country. A high proportion had no schooling; just over 50% in India and Ghana, and about 32% in China; about 3% had university level education in all three countries.

Table 2 shows the weighted prevalence of self-reported unmet need for oral health among adults aged 50 years and over in China, Ghana, and India. Prevalence was highest in Ghana (80%) followed by India (62%) and then China (60%). In China the prevalence of unmet need was higher in females (52% versus 48% in males), in Ghana the prevalence was higher in males (55% versus 45% in females) and in India the prevalence was approximately equal for males and females. Unmet need was more prevalent in rural areas in all three countries. In China the prevalence was 71% in rural areas and 29% in urban areas, in Ghana the prevalence was 66% in rural areas and 34% in urban areas, and in India the prevalence was 73% in rural areas and 27% in urban areas.

Unadjusted logistic regression models for household wealth and education are presented in Table 3. In China, respondents in the poorest household wealth group were significantly almost three times more likely to report unmet need, compared with those in the richest household group (OR 2.85: 95% CI: 1.90–4.27). In Ghana respondents in the poorest household wealth group were significantly three times more likely (OR 2.96: 95% CI: 1.20–7.23) to report unmet need. In India respondents in the poorest household wealth group were 40% more likely to report unmet need (OR 1.44: 95% CI: 0.83–2.50) although this estimate was not statistically significant.

The least educated in China were significantly six times more likely to report unmet need (OR 6.06: 95% CI: 2.27–16.17), and over six times more likely in Ghana (OR 6.39: 95% CI: 2.31–17.71). In India the least educated were almost five and a half times more likely to report unmet need (OR 5.40: 95% CI: 1.53–19.05).

Age was significant and inversely associated with self-reported unmet need in China and India, rural residence was significantly associated with unmet need in China as was “bad” self-reported health.

The adjusted logistic regression analysis for household wealth is shown in Table 4. In China, the household wealth variable attenuated to non-significance after controlling for age, sex, area of residence, marital status, work status, and self-rated health. In Ghana however, the association remained significant after controlling for potential confounders; those in the poorest household wealth group were almost three times more likely to report unmet need compared with those in the richest household wealth group (OR 2.84: 95% CI: 1.14–7.11). Household wealth was not significant in India.

Sex was not significantly associated with self-reported unmet need except in India where males were 60% more likely to report unmet need than females (OR 1.61: 95% CI 1.02–2.54). Sex was not significant in either China, Ghana or India in the unadjusted analyses (See Table 3). In China those in the 80+ year age group were almost three and half times more likely to report unmet need (OR 3.41: 95% CI 1.83–6.35) compared with those in the 50–59 year age group. In India those in the 80+ year age group were almost three times more likely to report unmet need compared with the 50–59 year age group (OR 2.89: 95% CI 1.39–5.98). However, the age gradient was not statistically significant in Ghana. Work status was statistically significant only in India. Compared with current workers, non-current workers were 50% more likely to report unmet need (OR 1.51: 95% CI 1.01–2.25) and those who had never worked were 75% more likely to report unmet need (OR 1.75: 95% CI 1.01–3.05).

The adjusted logistic regression analysis for education is shown in Table 5. In China education attenuated but remained significant, whereby the least educated were three times more likely to report unmet need compared with the most educated (OR 2.87: 95% CI 1.04–7.89) after controlling for age, sex, area of residence, marital status, work status, and self-rated health. In Ghana the odds of unmet need were over eight times higher in the no schooling group compared with the university group after controlling for the confounders, although the wide confidence interval indicates low precision (OR 8.26: 95% CI 2.81–24.31). In India those who identified in the no schooling group were seven times more likely to report unmet need compared with those with university education (OR 7.05: 95% CI 2.14–23.19) after controlling for age, sex, area of residence, marital status, work status, and self-rated health.

In this education model sex was again significantly associated with self-reported unmet need only in India where males were twice as likely to report unmet need as females (OR 2.11: 95% CI 1.27–3.50). In China those who identified in the 80+ year age group were almost three times more likely to report unmet need (OR 2.76: 95% CI 1.50–5.07) compared with those in the 50–59 year age group. In India those in the 80+ year age group were two and a half times more likely to report unmet need compared with the 50–59 year age group (OR 2.53: 95% CI 1.20–5.30). As in the wealth model, the age gradient was not significant in Ghana. Work status was again only statistically significant in India. Compared with current workers, those who had never worked were about 70% more likely to report unmet need (OR 1.69:95% CI 1.01–2.83).

Table 6 shows unadjusted RIIs for household wealth and education. In China the relative risks of self-reported unmet need among adults in the poorest household wealth group were significantly 60% higher than those in the richest wealth group (RR 1.61: 95% CI: 1.32–1.96). In Ghana the unadjusted relative risks were significant and 30% higher with the same comparison (RR 1.29: 95% CI: 1.04–1.61). The unadjusted RII was not statistically significant for household wealth in India.

The unadjusted relative risks of self-reported unmet need in the least educated group were significantly twice that of the highest educated group in China (RR 2.05: 95% CI: 1.65–2.56) and significantly one third higher in Ghana (RR 1.33: 95% CI: 1.09–1.61).

Table 7 gives adjusted RIIs for household wealth and education, after controlling for possible confounding effects by age, sex, area of residence, marital status, work status, and self-rated health. In this adjusted analysis the RII for household wealth was statistically significant only in Ghana (p < 0.05) where the relative risk of unmet need among adults in the poorest compared with the richest household wealth groups was almost 30% higher (RR 1.28: 95% CI 1.04–1.56).

The adjusted relative risks of self-reported unmet need among the least educated compared with the highest educated group, were significantly almost 50% higher (RR 1.49: 95% CI: 1.18–1.88) in China, significantly almost 40% higher in Ghana (RR 1.37: 95% CI: 1.12–1.67) and just over 50% higher in India (RR 1.53: 95% CI: 1.18–1.97).

Comparing the adjusted and unadjusted models, we see positive attenuation in China and negative attenuation in India. The net effects of older age, rural residence, never having worked and having bad self-rated health contributed to the education inequality in China, (Table 3). However, in India the net effects of the confounders offset the inequality because when they are held constant (in the adjusted analysis) the inequality increased to significance. This is also attributed to interaction between sex and education in the India sample. Men were generally more highly educated than women, and education is negatively associated unmet need, yet men were more likely to report unmet need for oral health care. In India therefore, association between education and unmet need was modified by sex.

The study has a number of strengths. In addition to extending the public health literature on oral health inequalities, the analyses contribute evidence of self-reported unmet need for oral health care by adults aged 50 and over in three contrasting LMICs. The data were captured and measured using validated WHO survey instruments. Survey questions pertaining to self-reported unmet need are consistent with those used by other international studies including the Survey on Health, Ageing and Retirement and the European Union Survey of Income and Living Conditions (22). Household wealth, as measured here using principal component analysis, is stable indicator of socioeconomic position in LMICs [57]. A further strength is the use of the RII which is a recognised robust measure of socioeconomic status in populations. Finally, given the global debates about universal health coverage and financing, these results can provide a basis for discussion about the need for publicly funded oral health services [58].

However, there are also some limitations. This was a cross-sectional analysis and causation cannot be claimed. Although WHO-SAGE used a stratified multi-stage sampling design to ensure that the country samples were representative of their national populations, the country samples were conditioned according to responses to two particular questions on oral health. We therefore acknowledge the possibility of sampling bias in the way in which unmet need was interpreted by respondents. The poor for example, may not be aware of their oral health needs because they have more important priorities such as obtaining and providing sufficient food and clean water for their families. Although our analyses showed for example that women in India reported less unmet need than men, we cannot infer from this that women’s oral health needs were met.

The analyses focused on curative oral health services in older adults. This was necessary because of data availability however we acknowledge that there is also the need for preventive interventions and services and that oral health applies across all age groups. This topic, as well as gender issues mentioned above, could be the subject of future research using qualitative as well as quantitative methodologies.

There are many factors associated with the measurement of oral health in LMICs and we were unable to cover them all. Psychosocial determinants such as social networks, social capital, self-efficacy, the environment, cultural attitudes and perceptions and also structural factors including physical infrastructure and supply of health services, medical insurance and distance to available oral health facilities, are some areas of follow up for future research.

The numbers of older adults will continue to increase in all parts of the world and LMICs are no exception [3]. The findings suggest questions for policy makers such as: How will unmet need for oral health care be met? What strategies are currently available? Given what is known about the links between NCDs, risk factors and oral health, what possible joint contributions can oral health and public health professionals make? [59]. How will preventive services and oral health education be integrated with curative services? How can taboos and myths about oral health be changed?

This study establishes the importance of education in seeking oral health care and supports the case for integrating oral health into general health promotion and education programmes in LMICs. Lessons should be learnt from the HICs where the take-up of health promotion messages has been greater among the more highly educated. New ways of targeting less well educated groups in LMICs with health promotion messages, are needed.

There is now more than ever a need to integrate oral health into public health policies [7, 60, 61]. New policies must be backed by sufficient human and other resources so that unmet need can be firstly recognised, and secondly, met through skilled service providers. The three countries in this study are all at different stages of development with regard to public oral health and they all face different challenges with regard to identifying and meeting the needs of their populations. Resources must be committed to oral health care infrastructure, education, data collection, the training and location of oral health professionals and oral health promotion. The goals of universal health coverage must explicitly include oral health. Political commitment and support is needed in all countries. This resonates with issues raised in the literature regarding the need for serious policy commitment to dentistry and oral health worldwide [62].