Decision Trees; Sensitivity Analysis, Gambia, Malaria Treatment

Introduction
Malaria is one of the biggest causes of infant mortality in Africa, where it accounts for about 4% of infant deaths and 1% of those aged 1-4 (Greenwood et al, 1987), particularly in the Gambia, which is recognised as one of the poorest countries in the world. (Harpham, 1996).  As a result the country is constantly looking for low cost initiatives that will reduce these figures and are simple and easy to implement.

However, before an initiative can be endorsed and put in place it is wise to consider all potential approaches through a structured process that takes account of all variables that could be impacted by the implementation.  (Weinstein et al, 2003).

One approach to malaria control is the introduction of a screening programmes: full screening, or partial screening only to the ceiling to prevent the entrance of the mosquitoes to the property.

This report covers the decision making process undergone to assess the best type of screening intervention to implement, utilising decision trees, and sensitivity analysis on a recent study into screening approaches.

Methodology
Kirby et al, (2009) describe a randomised control trial of 500 houses in a town in the Gambia, which currently has a low use of insecticide treated bednets.   These bednets are a common feature of malaria treatment in the Gambia (D’Alessandro et al, 1995), however their usage is dependent upon location and user participation in the programmes.  In the study houses were randomly assigned to receive full screening, screened ceilings, or no screening (control) but none of the screens were treated with insectide.  Assessment of efficacy included number of female anopheles gambia sesu lato mosquitoest trapped per night as a primary endpoint, with secondary endpoints of anemia frequency, and parasitaemia at the end of the transmission season.  

In order to assess the suitability of expanding the trials to other areas in the Gambia a decision tree was created (See Fig. 1).  In addition, the results from the study were subjected to a sensitivity analysis.  The choice of these two process was firstly because using a decision tree process tree allows for any study, but particularly those in the field of health care, to be assessed against potential outcomes and unconsidered or uncontrolled for side effects of the original study, to be assessed for it meaningfulness for dissemination outside the confines of a controlled study. (Cuijpers et al, 2005).

Secondly, in the case of a sensitivity analysis, the application of a one-way sensitivity analysis shows the impact of one small change in some of the study values or research results which will inform the success or otherwise of a wider application of the proposed intervention.  Changing one parameter at a time means that a variety of values can be assessed to provide information about the potential successful outcomes or otherwise of the proposed intervention to enable consideration of its economic, health and other benefits in the proposed region. (Briggs, 1994).

In conducting both a sensitivity analysis and formulating a decision tree it will be possible to analyse the study from the perspective of looking at the evidence base for the effectiveness of the various screening approaches, both in terms of economies and as a health benefit.   In particular using a decision tree will raise questions regarding the benefits and side effects of applying an evidence based approach to an ongoing medical problem. (Lifford & Braunholz, 1998) which will assist in making the final decision regarding which screening intervention to implement in the Gambia.

The application of both of these process will give a much clearer picture of which is the most appropriate screening intervention and clarify the evidence base for the health approach.  Whilst each one will provide a level of information, it is the combination of the two approaches that provides the most effective analysis and responds to the question of whether the evidence base for the screening interventions is sufficient to warrant its implementation. (Eddy, 2005) recognised that in the area of evidence based medicine a multi-analysis approach is the most effective way to consider research studies and their potential application in the field.

Decision Tree Process
As noted, the decision tree process allows the analyser to raise and answer questions about possible outcomes and their potential consequences, in this way it is possible to consider the effectiveness in both health terms and cost terms of implementing a particular practice. (De Ville, 2006).  In the context of looking at evidence based medicine, and in particular at potential interventions that will reduce the incidence of disease in a particular area, it provides a graphical illustration, which by placing values on particular outcomes, both actual and potential can be used to interpret results of studies and see whether they can be applied in a general and global way to answer question about the most effective treatment options.  (Timmreck, 2003).  It is important when constructing a decision tree to factor in human elements such as motivations and reactions, and particularly in the framework of healthcare, any potential or unconsidered side effects of the intervention.

Sensitivity Analysis
Once the decision tree has been created, subjecting available data to a sensitivity analysis provides further information on which is the most efficacious solution. A sensitivity analysis allows values to be placed on estimated risks of implementing a particular course of action, which can then be adjusted to see what impact small changes could have. (Saltelli, 2004). When applied to research into health care interventions this becomes very pertinent, particularly when dealing with different groups of individuals. Therefore, applying a sensitivity analysis to the results achieved in a study of screening interventions for malaria treatment will provide additional information that can be applied when making the final decision.

Model Development
In order to consider whether full, partial or no-screening would be an appropriate intervention, three decision trees were used, one to represent the reduction in a gambia, one for population motivation and one for presence of anaemia and paristaemia which are attached as Appendix 1.

By placing mathematical values on potential outcomes such as presence or otherwise of anaemia symptoms and presence of parasitaemia it is possible to calculate the probability of the effect of an intervention.

As noted above, placing values on a set of parameters allows for a sensitivity analysis to be created.  In order to assess how changing certain factors will impact on the decision of which screening intervention to select, a sensitivity analysis was created and this is shown below as Fig. 1.  The parameters selected for change and adjustment were costs of implementation per household, number of a gambia mosquitoes caught and percentage reduction in anaemia symptoms in children.   For best case scenario the cost parameters remained as the original, mid-case were raised by 5% and worst case by 15%.  This would allow for increases in cost in the netting. 

Scenario Summary
		Current Values:	Best case scenario	mid-case scenario	worst case
Changing Cells:
	$B$3	11.11	11.11	12.34	16.66
	$C$3	60%	60%	60%	60%
	$D$3	12%	12%	12%	12%
	$B$4	21.17	21.17	22.22	24.28
	$C$4	40%	40%	40%	40%
	$D$4	12%	12%	12%	12%
	$B$5	0	0	0	0
	$C$5	0	0	0	0
	$D$5	0	0	0	0
Result Cells:
	$D$2
	$A$3	Full screening	Full screening	Full screening	Full screening
	$B$3	11.11	11.11	12.34	16.66
	$C$3	60%	60%	60%	60%
	$D$3	12%	12%	12%	12%
	$A$4	Partial screening	Partial screening	Partial screening	Partial screening
	$B$4	21.17	21.17	22.22	24.28
	$C$4	40%	40%	40%	40%
	$D$4	12%	12%	12%	12%
	$A$5	No screening	No screening	No screening	No screening
	$B$5	0	0	0	0
	$C$5	0	0	0	0
	$D$5	0	0	0	0

Evidence Based Medicine
The place and role of evidence based health care is essentially to use apply current best research evidence to the prevention of disease and treatment of health disorders (Haynes & Haines, 1998).  In this case, the evidence from the Kirkby study suggests that the implementation of a full screening programme would prove beneficial in reducing some of the indicators of the development of malaria, such as the presence of anaemia and levels of parasitaema in infants.  However, when considering the evidence for any long term intervention it is also important to compare and contrast study results (Hunink & Glaziou, 2003) as well as consider wider implications.  These would include both benefits and side effects.  By referring back to the decision trees created to assess these factors, it can be seen that the benefits with full screening are reduction in presence of severe anaemia by some 8%, and reduction in high parasitaemia levels of some 7% when compared with no screening at all.  Whilst there are reductions in these levels with partial screening it is not as high.  The benefits of the implementation of full screening can also be seen in the motivation to participate by the population with 93% wishing to continue with full screening against only 45% wishing to retain partial screening. 

It is also necessary to consider any side effects.  As this intervention is not a treatment that requires direct medicine taking by the population, there are unlikely to be any physical negative side effects, it should be recognised that as individuals are all different and age, gender and overall health status can be impacted by any intervention these will need to be taken into account when measuring anaemia levels, parasitaemia levels and susceptibility to potential development of malaria in the future.

There are limitations to applying decision based analysis to evidence based medicine, (Naylor, 1995) specifically when looking at one intervention such as screening.  For example, the conditions which may impact the outcomes such as whether the houses are also using insecticided bed nets or other approaches need to be considered before a final decision is made. 

Results
The results of the analysis lead to a recommendation for full screening. It is however, prudent when looking at applying evidence based interventions the implications of implementation in the wider context. Sarasin (1999) noted that any trial is based around “average” groups and patient responses need to be considered. In this instance, where the implementation of a screening intervention will require support and co-operation from the population to be successful it should be registered that , 97% of the full screening group opted to retain the intervention following the trial, whilst 60% of those having been given partial screening requested the full screening, suggesting a high motivation for this intervention, which is borne out by the decision tree analysis

Conclusions
Whilst the overall impact of the implementation of a full screening programme would not reduce significantly reduce the instances of parasitaemia, the fact that it would reduce anaemia  in children, which is an indicator of potential susceptibility to malarial superinfection  (Sachs & Malaney, 2002) suggests that it would have a very strong benefit in the Gambia region. 

When compared to the results for partial or no screening, despite the cost implications which would be less than a full innoculation programme (Carter et al, 2000), the longer term benefits would amortise these costs by reducing health care needs as the numbers of children being struck down by malaria or anaemic symptoms would diminish over time.

In addition, by offering households an intervention that is practical and has benefits such as removing other pests and dusts, it increases their own sense of responsibility and ownership for their health care. Barratt and Brown (1996) noted that in the Gambia the development of maternal education and understanding of their responsibilities to the health of their children was an area that needed work and a programme such as this would encourage development and recognition of that need. 

The need to consider all of these factors when considering any research based evidence for longer term health care planning highlights the relevance of, the application of processes such as decision trees and sensitivity analysis can be a vital component in ensuring that all avenues are considered before an intervention from a research study is implemented in the wider environment.   Utilising these tools ensures that all potential issues are incorporated into the final decision, .  (Davies et al, 2003) i.e. the overall impact of the intervention, the target populations motivation for, and therefore likely adherence to, the intervention and any other potential side effects that may not have been accounted for in the study design or final research.

REFERENCES
Barratt H., & Browne, A., (1996), Health Hygiene and Maternal Education: Evidence from the Gambia, Social Science & Medicine
Volume 43, Issue 11, 1579-1590

Briggs, A., Sculpher, M., Brixton, M., (1994), Uncertainty in the Economic Evaluation of Health Care Technologies: The Role of Sensitivity Analysis , Health Economics, Volume 3, Issue 2, 95-1-4

Carter, R., Mendis, K.N., Roberts, D., Spatial Targeting of Interventions against Malaria, Bulletin of the World Health Organisatio, Vol. 78, No. 12, Geneva

Cuijpers, P., de Graaf, I., Bholmeijer, E., (2005), Adapting and Disseminating Effective Public Health Interventions in Another Country: towards a systematic approach,  European Journal of Public Health, Vol,. 15, No. 2, 166-169

D’Alessandro, U., Olaleye, B.O., McGuire, W., Thomson, M.C., Langerock, P., Bennett, S, Greenwood. B.M., A comparison of the efficacy of insecticide-treated and untreated bed nets in preventing malaria in Gambian children, Transactions of the Royal Society of Tropical Medicine and Hygiene
Volume 89, Issue 6, 596-598

Davies, R., Roderick, P, Raftery, J., (2003) The evaluation of disease prevention and treatment using simulation models European Journal of Operational Research, Volume 150, Issue 1, Pages 53-66

De Ville, B. (2006), Decision Trees for Business Planning and Data Mining,: Using SAS Data Miner, Cary

Eddy, D. (2005), Evidence Based Medicine: A Unified Approach, Health Affairs, Volume 24, Number 1, 9-17

Greenwood, B., Bradley, A., Greenwood P, Byass, K., Marsh, J.K, Tulloch, S., Oldfield, F., Hayes, R., (1987), Mortality and Morbidity from Malaria Among Children in a Rural Area of the Gambia, West Africa,
Transactions of the Royal Society of Tropical Medicine and Hygiene, Volume 81, Issue 3, Pages 478-486

Harpham, T., (1996), Poverty in the Gambia, Health & Place
Volume 2, Issue 1, March, Pages 45-49

Haynes, B, & Haines, A, (1998), Getting research findings into practice
Barriers and bridges to evidence based clinical practice, British Medical Journal, Vol 317, 273-276.

Hunink, M.G.M., & Glasziou, P.P., (2003), Decision Making in Health and Medicine, Intergrating Evidence and Values, (2nd Edition), Cambridge University Press

Lifford, R.J. & Braunholz, A., (1998), Getting Research Findings into Practice: Decision Analysis and the Implementation of Findings, British Medical Journal, 317, pages 405-409

Naylor, C. D., (1995), Grey zones of clinical practice: some limits to evidence-based medicine, The Lancet, Volume 345, Issue 8953, Pages 840-842

Sachs, J & Malaney, P, (2002), The Economic and Social Costs of Malaria, Nature, 680-686

Sarasin, F.P. (1999), Decision Analysis and the Implementation of Evidence Based Medicine, Quarterly Journal of Medicine, Vol 92, 669-671
Saltelli, A., (2004), Sensitivity Analysis in practice: A guide to assessing scientific models, John Wiley

Timmreck, T.C. (2003), Planning, Programme Development and Evaluation : A Handbook for Health, Jones and Bartlett

Weinstein, M.C., O’Brien, B., Hornberger, J., Jackson, J., Johannesson, McCabe, Luce, B.R. (2003), Principles of Good Practice for Decision Analytic Modelling in Health-Care Evaluation: Report of the ISPOR Task Force on Good Research Practices – Modeling Studies, Value in Health, Volume 6, No. 1, 9-17