Descriptive cross-sectional household surveys were conducted, and two-stage cluster sampling was used (Galway et al. 2012). A group of informants were identified through the randomised, stratified cluster sampling of households in the two areas. Stratification was based on the type of dwelling as identified in an aerial map. The households to be visited were selected by delineating and numbering equivalent in size areas on a Google map and then randomly selecting areas which appeared to consist chiefly of each type of dwelling, for example informal and formal, stand-alone houses and apartments. The number of dwelling type to be included was proportional to the census information (Housing Development Agency 2012). Excel Rand function was used to identify these clusters for inclusion. Data collection was begun at the corner dwelling on a randomly chosen block in each selected area, and continued along the street from there, visiting every alternate house. The informant was the head of the household or the eldest person available at the time of the visit. They were interviewed to identify people with disabilities; thus, proxy report was used to identify those with disability (Maart & Jelsma 2013).

The sample size was calculated using Epi-Info Stat Calc, Version 6. The 2001 census data were used (proposal submission for the research was submitted before the release of the 2011 census information), which reported the Oudtshoorn population to be 85 000 and that of Nyanga 58 723. As cluster sampling was used at the level of geographical area and household, a design effect of 1.5 was used (Skinner 1986). The estimated prevalence of disability was 5.2% (Statistics South Africa 2001). With these parameters, a sample of 2778 subjects in Oudtshoorn and 2753 in Nyanga was required (95% confidence intervals [CIs] and a 1.5% margin of error). Anticipating that the average household in this province consisted of 3.8 members in 2001 (Housing Development Agency 2012), the minimum number of households to be visited was approximately 900. To allow for non-responses, the final number of households to be visited was 1000 in each area.

Subjects of the surveys included all adults and children older than 5 years of age who were permanent residents in the identified household. Information was only gathered on older children as the WG Short Set is not appropriate for children under 5 years of age (Madans et al. 2004). Individuals who were visitors to the home or not permanent residents were excluded from the study.

A questionnaire was developed that included basic demographic details of all household members and the informants themselves (Maart & Jelsma 2013). This included income, education and employment questions. There were three sections concerned with demographic characteristics (e.g. age, education) social characteristics (e.g. household composition) and biomedical characteristics. The WG Short Set (six questions) was used to identify persons with disability (Madans et al. 2004). This instrument has been validated for use in South Africa, and has been used in the South African National Census since 2011 (Lehohla 2005; Schneider et al. 2009).

Home-based carers working in the respective communities were identified as research assistants to assist with data collection. The research assistants attended a training workshop on administering the questionnaire and a research supervisor from the community was appointed for quality assurance purposes (Maart & Jelsma 2013). Ethical approval for the study was obtained from the Human Research Ethics Committee at the University of Cape Town. Prior to conducting the interview, the study was explained to the informant and written and verbal consent was obtained. If for any reason, respondents in the identified household could not be interviewed, a further visit at an alternative time was attempted before exclusion of the household.

Descriptive statistics were used throughout. The percentage prevalence of the different levels of disability was calculated, with the numerator being the number of people reported to have scored ‘some difficulty’, ‘a lot of difficulty’ and ‘unable to do’ based on the WG Short Set Questions. The denominator was the total number of household members reported on by the informants. The demographic details and disability prevalence and severity were compared between the two areas, using Chi-square, t-test or the Mann–Whitney U tests, depending on the nature and distribution of the data. Confidence level was set at 95%.

The research assistants attended a training workshop on administering the questionnaire. A research supervisor from the community was appointed for quality assurance purposes. Ethical approval for the study was obtained from the Human Research Ethics Committee at the University of Cape Town (HREC number 098/2012) and all participants signed informed consent prior to participating in the interview.

Oudtshoorn informants were significantly older on average than those in Nyanga (p < 0.001). In Oudtshoorn, there were more female informants (50.7%) compared to Nyanga (48%). The mean age of the entire sample was 32.8 (SD 18.0) years and there was no significant difference in the mean age between the two areas. The average household size was greater and the care dependency ratio almost twice as high in Oudtshoorn (0.47) compared to Nyanga (0.25).

Some form of employment was reported by 38.8% of the total sample. A smaller proportion was employed in Oudtshoorn (28.0% compared to 49.9% in Nyanga). There were a large number of missing responses to the income question in both sites and the median income category of the total sample was R1000 – R2999 of those reporting. The Oudtshoorn respondents reported significantly less income than Nyanga respondents (p < 0.001).

In all, 645 individuals screened positive for having difficulty in any one of the domains, a prevalence of 9.7% (CI 9.0% – 10.4%).

The demographic information of the surveyed sample who did not have disability is compared to that of the people with disability (PWD) in Table 3.

People with disability tended to be older (p < 0.05) and there were more women (p = 0.001), higher unemployment (p < 0.001) and a lower income (p < 0.001) in this group compared to people without disabilities. People with disability in Oudtshoorn were older than their peers in Nyanga (p = 0.033) and had a lower median income (p = 0.017) (Table 4).

A significantly smaller proportion of respondents were identified as having a disability in Oudtshoorn (244, 6.8%, CI 5.9–7.6%) compared to Nyanga (401, 13.1%, CI 12.0% – 14.3%; Chi-sq = 77.8, p < 0.001) (Table 5, Figure 1). In addition, the CIs of the different severity levels did not overlap.

There was a significant association between age group and presence of disability in the overall sample. The Oudtshoorn sample showed a relatively higher proportion of elderly and a lower proportion of youth among those with disabilities, compared to Nyanga (Table 6).

The study samples were representative of their respective districts and the Western Cape (Statistics South Africa 2012), in that they demonstrated a higher proportion of women compared to men in the population, and that the mean ages found in the study are comparable to the census of 2011 (Statistics South Africa 2012). The study sample was impoverished relative to the Western Cape at large, with an approximate average annual household income (R36 000, approximately $3600), which is much lower than the average household annual income recorded for the Western Cape (R143 461/$9647.59) (Statistics South Africa 2012). The annual income described in the study would translate to approximately $30 per month, which is less than $1 per day; this is way below the upper-bound poverty line of R992.

People identified with disability, who were more likely to be female and were older on average, have a higher unemployment rate and lower income than the general sample. The trend of increased disability among women is well documented globally and locally (Ethgen et al. 2004; Palmer & Harley 2012; Statistics South Africa 2012; Trani & Bakshi 2008). The higher incidence of non-communicable diseases, associated co-morbidities and increased life expectancy among women could account for the higher prevalence of women screening positive for disability (Peeters et al. 2013). The mean age of those screening positive for disability was older than the general sample which is not unexpected (Kostanjsek et al. 2013). The lower income associated with disability in this study is in contrast to that of Jelsma et al. (2007), who reported the household income to be higher in South African households with a disabled member and suggested that the disability grant available to those with disability could be the reason for this (Jelsma et al. 2008). It appears that in an urban setting, disability is associated with added financial advantage, as Nyanga reported a much higher median income for both the overall sample and among PWDs.

The overall and site-specific prevalence rates in the study sample were higher than the average disability prevalence reported in the National Census (2011), despite the use of the same instrument to identify disability. As anticipated, the localised surveys revealed considerable differences between the two areas surveyed. Not only was there a variation in the overall prevalence of disability, the patterns differed with regard to levels of severity, ages and income of those with disabilities. While the overall rate of 9.7% is still lower than the WHO (2011) global estimate figure of 15%, the Oudtshoorn prevalence was 5% less than that of Nyanga. However, the prevalence of those who were ‘Unable to do’ although small was about three times higher in Oudtshoorn.

The disability prevalence in this study was found to be 4% – 5% for severe disability and 9% – 12% for moderate to mild functional limitations. These rates are similar to the World Health Survey (2013) (4% for severe and 16% for moderate to mild, respectively), and the Global Burden of Disease (2010) estimation (3% having extreme or severe disability and 15% having mild to moderate disability). The estimate for moderate to mild is somewhat lower (GBD 2010 Country Collaboration 2013; Ustun 2003). However, as hypothesised, the survey showed a much higher disability prevalence rate (9.7%) than reported in the South African census data (5.7%). The disability prevalence rates in both areas (Nyanga 13.1% and Oudtshoorn 6.8%) were higher than the Western Cape prevalence rate of 5.4% and comparable to less resourced provinces (Statistics South Africa 2014). These results may be partially explained by the continued deprivation and marginalisation experienced in disadvantaged communities in otherwise well-resourced provinces.

Apart from income and ethnic background, other aspects of the context and social environment are very different between the two areas. The burden of disability is borne by different groups in the two areas. In Nyanga, the mean age of those with disability was lower and 80% of those with disabilities were in the 15–59 year ‘Adult’ group. The increased prevalence of ‘some’ and ‘severe’ difficulties in this age group in Nyanga could be ascribed to the fact that Nyanga is described as one of the most violent neighbourhoods in the country, leading to high rates of trauma-related disability (South African Police Service). The high level of interpersonal violence is fuelled by rapid urbanisation and socioeconomic disparities which are much more evident in the urban context (Mont 2007b). These inequalities contribute greatly to the burden of disease and disability in South Africa (Norman et al. 2007). The extremely high population density of Nyanga (Table 1) (18 775 persons per square kilometre compared to only 27 persons in Oudtshoorn) may also be a factor. The impact of so many people living in a confined physical space is likely to manifest itself in pressure on facilities, lack of open spaces for recreation and invasion of personal space, all of which may contribute to the poor social climate evident in the high crime statistics (South African Police Service).

The difference in average age and household size between the sites could also be linked to urban migration. It is likely that those who become severely disabled often return to their families in rural areas, and this could explain why the proportion of people ‘unable to do’ is higher in Oudtshoorn (nearly three times as many) than in Nyanga. Another explanation could be lack of access to health services, which might be related to screening of health-related complications leading to more serious outcomes (Maart & Jelsma 2013). This, coupled with the higher care dependency ratio, indicates that caregivers in Oudtshoorn may well have difficulty in providing the required care to those with the most severe disability (Maart & Jelsma 2013). At the same time, those in Nyanga live in smaller households and they may not have someone who is able to aid and care. More research is thus required to adequately inform service delivery needs in the two areas.

Great emphasis has been placed on obtaining comparable data (Madans et al. 2004; Mont 2007a; World Health Organization 2001); however, the results of this study show that the simple reporting of only the prevalence of disability fails to capture the differences between communities. For services to be objectively prioritised, a more nuanced understanding of those presenting with disability is required. Although national or large-scale surveys such as the census can give general estimates of disability, these may not be useful for service planning. This study demonstrated that two areas, both of which are economically deprived, showed very different patterns of disability, both to one another and to provincial figures (Maart 2015). Locally specific information relating to persons with moderate to severe disability is of most relevance to policymakers and service providers, particularly in resource-constrained environments where there are competing priorities.

According to Helander (1992), it is reasonable to assume that persons with severe disabilities are dependent on others physically, economically, socially and psychologically, and would thus require a multidimensional approach to combat exclusion and facilitate their participation in society (Helander 1992). In both areas, our results indicate the need for continued financial support through disability grants and the creation of job opportunities, as PWDs appear to be less able to find employment and live in households with lower monthly income than those without disability.

The major weakness of the study was that the estimates of disability were based on the proxy reporting of informants (Maart 2015). It is well known that proxy reports often result in under-identification of subjective problems (Maart 2015), such as mental and emotional problems, as these are not as readily apparent as physical or sensory disabilities. It is also acknowledged that the prevalence of depression and other mental disorders is likely to be under-reported as it is not included in the WG Short Set. A further limitation is that children younger than 5 years of age were excluded from this study as the WG Short Set has not been validated for this age range, and this gives a skewed view of disability and services required (Madans et al. 2004).

The number of refusals in Nyanga is a limitation that should be investigated as this group could have added to the disability prevalence.

A further limitation could also be that the census data of 2001 was used to estimate the sample size.