Health policy support under extreme uncertainty: the case of cervical cancer in Cambodia

Abstract

Health policy support in least developing countries is needed to allocate scarce resources most efficiently and produce the maximum health at given resources. However, planners frequently face severe uncertainty of biological, economic and health service structures and parameters. This paper presents a model of cervical cancer (CUC) in Cambodia as an example of health policy support under extreme uncertainty. The projections are based on a multi-compartment dynamic policy model, specifically developed for CUC in Cambodia. The model simulates the demographic system and infections through sexual intercourse for 100 years. Data were taken from the literature and adjusted for Cambodia through interviews. CUC is an increasing problem in Cambodia and the number of deaths due to cervical cancer growths faster than the population. On average, Cambodia will lose some 5159 years of life per year due to CUC. From the alternative interventions against CUC, a “see-and-treat” approach based on VIA screening of women 30–49 years every 3 years seems to be most efficient. The results of the simulation indicate that the “see-and-treat” approach should be implemented in Cambodia. Even under strong changes of parameters and assumptions, this finding is robust. The model is currently being used in development planning. The example shows that health policy support is possible even under extreme uncertainty if the model builder employs a sufficient number of sensitivity analyses and scenarios.

Introduction

Health policy makers in developing countries are facing severe challenges in the attempt to cover a growing population with preventive and curative health care services. Infectious diseases (e.g., malaria, dengue) remain a severe problem while chronic-degenerative diseases (e.g., diabetes mellitus, cardio-vascular diseases and cancer) emerge. Policy makers have to allocate extremely scarce resources to different levels of health care (e.g., first-line facilities, district hospitals and national hospitals), to different regions and to different interventions against the respective diseases to achieve national health objectives most efficiently (Walt 2001). However, any allocation entails opportunity costs, i.e., inefficient allocation of public health care resources results in the loss of human life.

This situation calls for evidence-based policy making, i.e., decision makers, have to know the absolute number of cases of a disease today, and in the future, as well as the relative importance of this disease in comparison to other diseases (Flessa et al. 2011). Only in this way, they can decide whether this disease should be prevented and treated by public services and whether public funds should be utilized to subsidize these services (e.g., by waiving of user-fees). There is no doubt that an evidence-based policy support is of high importance in these countries. Seeing the high complexity of disease systems, modeling becomes a must to support health policy decisions (Tan-Torres and World Health Organization 2003).

However, this policy support has to be offered under uncertainty (Pitman et al. 2012). Firstly, there is uncertainty considering biological and medical facts. Modeling requires explicit statements about the relationship of certain parameters to each other, but much remains unknown. For instance, the natural reservoir of Ebola in Africa is still not perfectly certain (Pourrut et al. 2005) which challenges all policy advice for fighting this disease (Bente et al. 2009). In addition, many biological or medical parameters are uncertain and/or short-term variables, such as infectivity rates (Wawer et al. 2005).

Secondly, the vagueness of medical care structures and data is another source of uncertainty in disease modeling. For instance, predicting the cost of treating diabetes is only feasible if the clinical pathway and the prevalence rates are known. In many countries, there are no national guidelines in place for the majority of diseases.

Thirdly, many economic facts are unknown (e.g., cost per vaccination, cost per treatment unit, etc.). In particular, the time preference rate and the associated time horizon are of high relevance for any health economic analysis (Beutels et al. 2008). In the developed world, it is assumed that the market interest rate is an appropriate proxy of the societal time preference rate (Frederick et al. 2002), but for developing countries this is questionable as the majority of decision makers have a very short time horizon and a high time preference (Sauerborn et al. 1996; Wurthwein et al. 2001). In addition, most health economic analyses use a single human life as time horizon (Pitman et al. 2012). However, in developing countries with (at least in Asia) a rapidly increasing life expectancy, it might not be easy to assess what “life-time” actually means. For these countries in demographic, epidemiological and economic transition, equilibrium models which analyze merely steady states might be insufficient as, in particular, the transition period is of interest.

Finally, health policy support frequently requires some knowledge of social patterns, such as partnership behavior. It is obvious that a model of sexually transmitted diseases will require some knowledge of relationship patterns. At the same time, basic parameters might be non-existent in the literature, e.g., intercourse frequency (Gray et al. 2001).

There is no doubt that uncertainty about biological, medical care, economic as well as social structures and parameters exists in all settings. However, the uncertainty in developing countries is more severe. Firstly, existing data are highly unreliable. Until today, the most serious problem in any modeling of (health) economic interventions in developing countries is “the almost complete unreliability of what data are available and the absence of any recent data” (Lane 1984). Although much has been invested in health information systems, reliable data are still scarce in many settings.

This statement is, secondly, in particular true for interventions that have never been performed before in a certain country, i.e., if no clinical standards, no pathways and no experiences exist. Cancer, for instance, has not been treated in public facilities of many least developed countries (World Health Organization 2002). Introducing cancer treatment will require making decisions without any country-specific data. However, data from other settings might, thirdly, not be applicable to a specific country. For instance, it was observed that the onset of diabetes in Asia is some 10 years earlier than in Europe (Ramachandran et al. 2010) so that using data from Europe might be misleading in Asian countries. Frequently, we will find a situation where no local data are available and transferability of western data is questionable.

Finally, culture might impede estimating data. For instance, recording sexual habits is difficult all over the world, but in some countries it is a taboo (Schmitt 2004). Thus, not much is known about promiscuity and homosexuality in Asian countries which makes forecasting HIV/Aids for these regions difficult.

Consequently, we face a dilemma. On one hand side, modeling is absolutely necessary to support health policy making, on the other side it is under extreme uncertainty. Decision making under uncertainty has been described in general (Hey et al. 2010) and for health systems support (Briggs et al. 2012). In principle, we have to apply five measures which may reconcile the conflict (Pitman et al. 2012). Firstly, the model building has to put all emphasis on analyzing the existing literature (including existing models) to determine structures and parameters. Secondly, the model must be calibrated until it fits existing data (as far as available) as good as possible. This means that parameters are changed within a reasonable range until the simulated results are close to the real data (Karnon and Vanni 2011).

Thirdly, results must be questioned by a number of sensitivity analyses, i.e., parameters are changed and consequences are recorded (Sharif et al. 2012). Fourthly, the simulator must run different scenarios to see the consequences of structural and parameter changes. Finally, all results must be analyzed thoroughly and interpreted with great caution. With many different scenarios and sensitivity analyses, the scientist cannot provide one outcome parameter as the only true answer. Instead, he will provide insights and bands of cost-effectiveness that are robust against changes. In this way, he can provide policy makers with the evidence required for decision making.

In the following, we will present a case study exemplifying these principles. We have chosen the case of cervical cancer in Cambodia as it shows all characteristics of modeling under extreme uncertainty. We had to face this uncertainty when we gave consultancy to the Ministry of Health in Phnom Penh concerning the introduction of non-communicable diseases into the basic health care package of this country. For this purpose, the next section provides some background on cervical cancer and the situation in Cambodia. Afterwards, the system dynamics model is presented. The fourth section presents the results of different simulations. Finally, these results are discussed under the focus of health policy support under extreme uncertainty.