It is widely agreed that health services in developing countries need to be "scaled up" to achieve the Millennium Development Goals (MDGs). The MDGs, which were adopted in 2000 at the United Nations, set ambitious goals for reducing child and maternal mortality, combating HIV/AIDS and malaria, and achieving high levels of coverage for basic health services. New global health initiatives (such as the Global Fund to Fight AIDS, Tuberculosis, and Malaria (GFATM), the World Bank Multi-Country HIV/AID Program (MAP), the US President's Emergency Fund for AIDS Relief (PEFFAR), the GAVI Alliance, the Roll Back Malaria Partnership, and the Stop TB Partnership), and increased financial resources [1
] have raised expectations to deliver health programs at large scale. Although not explicitly defined by the global health initiatives, a working definition of scaling up has been proposed as "an ambition or process of expanding the coverage of health interventions" [2
Most of the recent emphasis on scaling up has focused on achieving high coverage rates of health services and reducing mortality, rather than the processes for how to scale up. The MDGs are identical for all countries, which in the case of childhood mortality, sets the target as a two-thirds reduction in child mortality rate between 1990 and 2015, the equivalent of an average annual decline of 4.3%. For the most part, the scaling up process is seen as the replication of specific health interventions (e.g. immunization, skilled birth attendance, integrated management of childhood illness, etc.) that have been shown to be cost-effective in a limited number of settings - usually a research setting or special projects conducted in a few countries. The intervention is expected to be delivered through a better-resourced and enlarged public health delivery system by replicating a similar package of interventions at more points of service delivery, often through model district health systems. HIV/AIDS programs are somewhat different in that they also enlist large numbers of non-governmental organizations (NGOs) for some of the interventions. The process of expanding coverage also involves the swift disbursement of funds, expanding partnerships, ensuring sustainable funding and promoting ownership, particularly at central levels [3
]. These assumptions have been used to estimate the costs of scaling up of several health services intended to achieve the MDGs [6
], and to estimate the human resources needed to provide them [7
]. Although in some cases adjustments are made for expected economies of scale, the expanding coverage of specific health interventions is largely expected to be independent of each other.
Unfortunately, many countries are not on track to achieve the MDG health goals by the end date of 2015 [8
], with one quarter to one half of all countries projected to not achieve their target levels for health services [9
]. Previous analysis of national trends in coverage of MDG-related health services where there exists sufficient data for trend analysis demonstrated that countries have very different rates of change of coverage [9
]. It was also found that changes in the rates of coverage of some health services are associated with the rates of change of other services (the services with sufficient data to assess annual rates of change include childhood immunization coverage, skilled birth attendance, tuberculosis treatment completion, and tuberculosis case detection). Using World Development Indicators data and the same multi-level statistical models as were used to assess changes in health services [9
], we also analyzed the annual rates of change in under five mortality for each low and middle income country between 1990 to 2009 (Figure ). Each line represents one country, and suggests that rates of change are highly divergent from one country to the next, and that the concept of an average country or average rate of change does not represent past experience. Figure uses the same data to plot a country's under five mortality rate in 1990 against its subsequent annual rate of change, demonstrating that many countries are not reaching the target level of 4.3% reduction per year, and that there are very few clusters of countries with similar rates of change and starting points. These results suggest that even if common sets of health interventions and common health goals are being pursued by the MDGs, the rates at which they are being scaled up is quite different, and with some countries even losing coverage and increasing mortality.
Figure 1 Trends in Under-Five Mortality for all Low and Middle Income Countries, 1990-2009. Note: Each line represents one country's trend, based on a multi-level model with random intercepts and random slopes for each country  Source: World Development Indicators (more ...)
Figure 2 Comparison of National Under Five Mortality Rates in 1990 and Subsequent Annual Rates of Change for Low and Middle Income Countries. Note: Each dot represents one country, with the trend based on a multi-level model with random intercepts and random slopes (more ...)
Notably absent from much of the discussion around scaling up to reach MDG targets has been the logic models or theories of change that can guide practice and research [12
], though a number of recent papers have examined why scaling up is not happening at the targeted rates. A recent review of the literature on scaling up in international health identified a number of common constraining factors, including the lack of absorptive capacity, weak health systems, human resource limitations, and high costs [13
]. On the other hand, strong leadership and management, realistic financing, and technical innovation are believed to be common characteristics of successful large scale health programs [13
]. However, there has been little attention paid to understanding why these factors occur and how to change them in ways that would guide practice for scaling up [12
]. Recent publications that have addressed how scaling up occurs have concluded that scaling up processes are complex [15
], with changes in political context and program management factors being major sources of variation in how scaling up occurs [16
]. There have also been warnings against the over-reliance on "gold standard" evidence on intervention cost-effectiveness as basis for policy and implementation choices, as there are limitations on how relevant they are to what will happen in a particular country [18
Because of the continued interest in scaling up health services, and the limited application of theoretical perspectives on scaling up in the current efforts to achieve the MDGs, it is a good time to examine what theoretical models have been used in scaling up for health in low and middle income countries. The purpose of this paper is to identify the theoretical frameworks that have been used to understand the issue of scaling up in the health sector in developing countries, and to identify how they inform practical questions of planning, implementation, monitoring, and evaluation for scale. It is hoped that by examining the theoretical approaches and lessons learned from them, we can gain insights for how to achieve the MDGs.