Findings from this study show that the prevalence of dependence increased sharply with increasing age, was higher among women than men, and among those with least education. Overall prevalence for those aged 65 years and over varied among 10/66 sites, from 2.9% in urban India to 15.7% in urban China. In most other sites, the indirectly standardized prevalence of dependence was one half to three quarters of that in the USA National Long Term Care Survey. The tendency towards a lower prevalence in rural sites in Latin America and China and in rural and urban India was accentuated after directly standardizing for demographic and chronic disease status. Dementia emerged as by far the leading independent chronic disease contributor to dependence. Limb weakness, stroke, depression, eyesight problems and arthritis made more modest contributions.
The main strengths of our study are the standardised design and assessment procedures carried out in representative catchment area samples across seven MIC, providing reliable and harmonized data on a wide range of cognitive, mental and physical morbidity among older people. This facilitates international cross-cultural comparisons regarding the prevalence and correlates of dependence. The main weakness was that dependence was ascertained using a semi-structured interview, and the rating of level of dependence was somewhat subjective. We chose this pragmatic approach, in the absence of previous research in LMIC, given the difficulties of developing a more structured assessment with demonstrable validity across many different countries and cultures. Other studies have inferred dependence from limitations in core activities of daily living, usually ascertained from the participant. Our approach was more direct, and the ascertainment of needs for care from the care provider, rather than the care recipient may have avoided under-reporting due to social desirability or cognitive impairment. Data on the inter-rater reliability of our assessment would have been valuable. Future cross-cultural comparisons would be assisted by a clearer operational definition of the construct. We did not cover the effects of cancer, endocrine disorders, genitor-urinary conditions and oral conditions on dependence, but these were likely to have been minor . More importantly, different conditions were ascertained with different levels of rigour; dementia, depression and hypertension by clinical assessment, but heart disease and stroke by self-report of medical diagnosis and visual and hearing impairment by self-reported impairment. Assuming random misclassification, this may have tended to reduce the size of any observed effect on dependence towards the null. The problem of self-report has been discussed by Amartya Sen who proposed that 'people in states that provide more education and better health facilities are in a better position to diagnose and perceive their own morbidities than are the people in less advantaged states, where there is less awareness of treatable conditions (to be distinguished from "natural" states of being)' . Finally, our data are cross-sectional. Therefore we cannot infer causality from the observed associations between health conditions and dependence. Some associations might have been inflated by reverse causality, thus depression can be a consequence as well as a cause of dependence [5, 33]. Information bias may also have occurred, since interviewers' ratings of the informant's account of needs for care may have been influenced by knowledge of the participant's health status.
Our estimates of the crude prevalence of dependence among those aged 65 years and over in MIC are generally lower than those reported in previous population-based studies of older people in high income countries; in England and Wales  (15.7% with significant disability among whom 86% had dependency needs), Scotland  (15% with short interval dependence), Spain  (15.5% with dependence in one or more of seven ADLs), France  (12.4% confined to home or bed) and the USA National Long Term Care Survey  (17.1% disabled in one or more activities of daily living, or living in a care home). Indirect standardisation, using the age-specific prevalences reported in the last of these studies confirmed this impression for all sites other than urban China. A relatively lower age-specific prevalence of dependence in MIC may be explained by a lower prevalence of chronic disease. Alternatively, given that prevalence is the product of incidence and duration, it may be that survival in a state of dependence is much shorter in MIC settings. We found, after standardizing for the main chronic disease correlates of dependence, that prevalence was lower in rural sites in Latin America and China and in rural and urban sites in India. This suggests another possible explanation. In these traditional and less developed settings, where most older people live with their children and are routinely provided with support for both core and instrumental activities of daily living, it may be difficult to identify 'the need for frequent human help or care beyond that habitually required by a healthy adult'. In Egypt, urbanisation has contributed to a growing awareness of unmet needs for care among older people; poor immigrant families living in slum districts need to work to maximise household income, leaving dependent older relatives without assistance .
We found, consistently across a wide range of MIC settings, that dementia is by far the largest contributor to dependence in the older population. This finding is analogous to that on the correlates of disability from the same 10/66 group surveys  although the effect sizes and population attributable prevalence fractions for the association with dementia are much larger for dependence than for disability. Other neurological and neuropsychiatric conditions - limb paralysis or weakness, stroke and depression - featured prominently in the list of leading contributors to dependence. This pattern of findings is entirely consistent with a large body of literature from high income countries. In a cohort study of Medicare recipients in the USA the onset of dementia at 12 months was strongly associated with the onset of dependence by 36 months (adjusted OR 7.5), low body mass index (OR 6.1), psychiatric disorder (OR 4.5), stroke (OR 2.5) and obesity (OR 2.1) also being independently associated. The onset of coronary heart disease, cancer, hypertension, lung disease, diabetes and hip fracture did not predict dependence . Similar findings were reported from a three year follow-up of a population-based cohort study in Sweden . Predictors of institutionalisation were very similar in a meta-analysis of 77 longitudinal community-based studies from the USA . Cognitive impairment was the strongest predictor of institutionalisation (RR 2.54), the increased risks associated with cancer (RR 1.15), hypertension (RR 1.04) and diabetes (RR 1.35) being modest in comparison; there were no associations observed with cardiovascular disease, arthritis, or lung disease. In Sweden, the population attributable fraction for the association between dementia and incident institutionalisation was 61% .
The gradient in the prevalence of dependence among older people, between HIC and MIC, and between urban and rural and least and more developed sites in our surveys suggests the potential for a substantial shift in the global profile of dependence, occurring mainly in low and middle income countries, and linked both to rapid demographic ageing and the health transition. There will be unprecedentedly rapid increases in the numbers of older people, and the prevalence of chronic diseases amongst them. Dependence, a consequence of chronic disease disability, will increasingly come to dominate the health and social care agendas in these countries. The proportions of dependent persons who are aged 60 and over will increase between 2000 and 2050, from 21% to 30% in sub-Saharan Africa, from 23% to 44% in India, from 23% to 47% in Latin America, from 30% to 60% in China, compared with from 45% to 61% in HIC . Over this period numbers of dependent older people are forecast to quadruple in most LMIC, while numbers of dependent younger people remain relatively stable. Therefore, in all world regions dependence is rapidly becoming a problem associated with ageing processes, particularly chronic disease morbidity. In the USA, compression of morbidity  was observed to have occurred in successive cohorts enrolled into the American's Changing Lives (ACL) study . Thus, at least for those with higher levels of education, increases in life expectancy comprised additional years of healthy life, rather than years lived with disability. For the least educated the pattern of a linear decline in health and functional status persisted in successive cohorts. As the demographic and health transitions impact on LMIC, the extent to which the chronic disease epidemics are prevented and controlled, and the extent to which improvements in public health and clinical care are equitably distributed will have a major impact on future long-term care requirements, and the attendant societal costs. There is an urgent need for these trends to be monitored in LMIC, using similar methodologies to the ACL studies.
Preventive interventions targeting older dependent people should be prioritised, mindful that according to the compression of morbidity hypothesis, healthy ageing, and healthy lifestyles may postpone the onset of chronic ill health and disability in the final years of life. Regardless of the success of such initiatives, numbers of dependent older people will increase markedly in the coming decades particularly in MIC, and the dependency ratio (the ratio of the dependent population to the 'working-age' population) is also set to increase from 8% to 14% in China and from 9% to 12% in India, compared with from 7% to 10% in developed countries . Under the most pessimistic scenario, by 2050 the dependency ratio will have reached 20% in China. It is therefore imperative that LMIC make policies and plans for the future provision and financing of long-term care . Some expansion of the care home sector from a very low base seems inevitable, regardless of government and cultural disapproval . This process needs to be monitored, and the emerging industry needs to be regulated for quality of care. As a counterpoint, informal care can be incentivised through the provision of non-means tested pensions for older people, and compensatory disability and caregiver benefits [4, 48]. Most importantly, family caregivers need to be supported in their role, a task currently neglected by community healthcare services . Initial findings from randomised controlled trials of our 'Helping Carers to Care' intervention, for caregivers of people with dementia in Moscow  and India  suggest considerable potential benefits for caregiver strain. Implementation of such interventions and policies will be a challenge in resource-poor LMIC. The World Health Organisation through its Mental Health Gap Action Programme (mhGAP) is finalising the development of evidence-based guidelines for the treatment of mental and neurological disorders by non-specialist health care workers in LMIC, providing for the first time a feasible and cost-effective model for scaling up services in these regions . Dementia and depression are two of the eight priority disorders, and the teams involved in developing the guidelines have recently published outlines of packages of care for these conditions [53, 54].