Applying the net-benefit framework for assessing cost-effectiveness of interventions towards universal health coverage
© Hounton and Newlands; licensee BioMed Central Ltd. 2012
Received: 22 December 2011
Accepted: 28 June 2012
Published: 16 July 2012
In assessing the cost-effectiveness of an intervention, the interpretation and handling of uncertainties of the traditional summary measure, the Incremental Cost Effectiveness Ratio (ICER), can be problematic. This is particularly the case with strategies towards universal health coverage in which the decision makers are typically concerned with coverage and equity issues. We explored the feasibility and relative advantages of the net-benefit framework (NBF) (compared to the more traditional Incremental Cost-Effectiveness Ratio, ICER) in presenting results of cost-effectiveness analysis of a community based health insurance (CBHI) scheme in Nouna, a rural district of Burkina Faso. Data were collected from April to December 2007 from Nouna’s longitudinal Demographic Surveillance System on utilization of health services, membership of the CBHI, covariates, and CBHI costs. The incremental cost of a 1 increase in utilization of health services by household members of the CBHI was 433,000 XOF ($1000 approximately). The incremental cost varies significantly by covariates. The probability of the CBHI achieving a 1% increase in utilization of health services, when the ceiling ratio is $1,000, is barely 30% for households in Nouna villages compared to 90% for households in Nouna town. Compared to the ICER, the NBF provides more useful information for policy making.
Context and participants
The Nouna community based health insurance scheme (CBHI) was launched in 2004 and was developed by the Nouna Health Research Centre as an operational research project to study how to improve community access and uptake of health services and how to meet the need of the poor within Nouna health district. The intervention has been extensively described in the literature [9–13]. It is a voluntary community health insurance scheme which aims to reduce financial barriers (out-of-pocket payments) and improve quality of care, thus improving access and uptake of health care. The alternative intervention will be the status quo (no enrolment in the CBHI).
Data collection and analysis
Data were extracted from the longitudinal Nouna demographic and surveillance site on membership of the Nouna CBHI, utilization of health services, the average distance from village to health centre, assets ownership, age and education level of the head of household. In order to generate household-level costs of the CBHI scheme from a societal perspective we added, for every household member of the scheme, the household costs (enrolment fees and premium) to the average cost of enrolling in the Nouna CBHI scheme from the health system perspective. The latter was obtained by dividing the 2007 annual costs of running the Nouna CBHI scheme by the number of households, members (370) in 2007. For households which are not members of the scheme, there is obviously no cost incurred for membership fees. However, these households have to meet the cost of the utilization of health services out of pocket. We use the estimated cost of the benefit package of the Nouna CBHI which was 9630 West Africa francs (XOF), equivalent to approximately $20) in 2004 . We computed ICER for extra additional utilization of health services, and cost effectiveness acceptability curves (CEAC) to illustrate the decision rule of cost effectiveness of the intervention.
The ICER decision rule is that if the estimated ICER lies below the ceiling ratio, which represents the maximum decision makers are willing to pay for an incremental unit of the measure of effectiveness, then the intervention concerned is deemed cost-effective. By varying the ceiling ratio, the varying probability that the intervention is cost-effective can be identified. The CEAC shows the probability of the intervention being cost-effective for all potential values of the ceiling ratio. Unlike the ICER approach, the CEAC can also be employed to obtain a confidence interval of cost-effectiveness. It also avoids the problems of interpretation of a negative ICER [14, 15]. There isa straightforward graphical representation and interpretation that a new treatment is not cost effective . In this paper, the likelihood values that the Nouna CBHI is cost-effective compared to the status quo were obtained using the p-values on the Nouna CBHI dummy when running an Ordinary Least Square (OLS) regression. The p-values are 2-sided p-values; however only one-sided probability is needed to test whether the incremental net-benefit is positive (Nouna CBHI is preferred) or negative (the status quo is preferred). The one–sided p-values were thus obtained by dividing the 2-sided p-values by 2. For negative incremental net-benefits the probabilities that Nouna CBHI is preferred equals the one sided p-values, and for positive incremental net-benefits, the probabilities that the Nouna CBHI is preferred equals 1 minus the one sided p-values (Hoch JS et al, 2006). A major strength of this technique when it comes to resource allocation is that for a given budget, one can model the different probabilities that the Nouna CBHI is preferred to the status quo.
However, whilst this technique could be sufficient in clinical care decisions about the choice of preferred medication, technology, or screening exam, in the public health field, a decision maker is often concerned about issues beyond the optimality of one intervention over another, especially with issues of equity. This is where the net-benefit framework could potentially be very useful in assessing the effect of significant determinants on the marginal cost-effectiveness of a universal health coverage intervention such as the Nouna community based health insurance.
Given access to health care is influenced by major determinants (such as the distance to health facilities, education, or assets ownership), a net-benefit framework, applied to the cost effectiveness of the Nouna CBHI, could be effective through the joint probability distribution in identifying the most important determinants that affect the cost-effectiveness results. The net-benefit framework employs linear regression techniques, and to date, has been most often used alongside clinical trials of health care regimens or technology devices [2–5]. Thus, it has the potential even for observational studies with patient-level effect and cost data, for the better presentation and interpretation of cost-effectiveness results and better evidence based decision making.
The traditional equation ΔC/ΔE (ICER) can be re-arranged by multiplying each arm of the equation by ΔE. The result is ΔC = ΔE * ICER and for any ceiling ratio Ro, ΔC = ΔE * Ro. Thus, a net-benefit statistic can be computed as follows: ΔE * Ro − ΔC = ΔNB. We computed for each observation (household) in the household survey an individual net-benefit statistic. The expression of an individual net-benefit NBi = ΔEi * Ro − ΔCi is similar to a traditional linear regression equation Y = α + δXi + εi where Y is the dependent variable, α is the intercept, δ the coefficient on an explanatory variable (continuous variable or dummy variable taking the value 1 for a positive outcome and 0 for a negative outcome for example) and εi is the standard error. Thus, for the Nouna community based health insurance scheme, the household net-benefit could be modeled as NBi = α + δCBHIi + ε i where NBi is the net-benefit for each subject (or household), α is the intercept, CBHIi, is the intervention (taking the value zero if a household is not a member of the scheme and 1 for a member), δti, is the incremental net benefit and εi is the standard error. The interpretation is straightforward and when this difference is greater than zero, it means that the incremental cost for one additional unit of effectiveness (in this case utilization of health services) is below the Ro (the maximum the provider is willing to pay). The CBHI will be deemed cost-effective in relation to the status quo. Similarly, if the coefficient is negative, then the incremental cost for one additional unit of effectiveness is above the Ro and the status quo will be deemed cost-effective.
Relative advantages of net-benefit framework and incremental cost-effectiveness ratio for presenting and interpreting results of cost-effectiveness analysis
Relative advantage criteria
Standard Incremental Cost Effectiveness Ratio (ICER)
Net Benefit Framework
Type of analysis
Descriptive analysis, and stratified analysis (by important covariates)
Regression analysis, and joint probability distribution with important covariates
Requirement of contextually relevant threshold (ceiling ratio)
Yes, to assess if intervention is good value for money
No, hypothetical ceiling ratios can be plotted and probabilities of cost effectiveness calculated
Adjustment to covariates (important sub groups)
Variability explained by covariates
Relative advantages for interpretation
Simple point estimate, greater or lower than a ceiling ratio
Graphical presentation; illustration of alternative scenarios with different ceiling ratios
The study was approved by the ethical review board of Nouna Health Research Centre.
Descriptive analysis of the study populations
Descriptive characteristics of populations by enrolment status (from household survey, 1504 household, 2007)
Use of health services
- Did not use
- Have used
- At least primary school level
- Nouna town
- Nouna villages
- Most poor
- Second quartile
- Third quintile
- Fourth quintile
- Least poor
Standard cost-effectiveness analysis
Sample statistics from the economic evaluation of the Nouna CBHI, data with net-benefit, household survey, 2007, Nouna districts Burkina Faso
(Not members N = 982 )
(Members N = 364)
Effect difference (%)
Place of residence
Nouna villages group
(%) Effect difference
Nouna town group
(%) Effect difference
Values of ceiling ratio (R)
R = 0
- 60623 (193)
- 58535.5 (507)
- 64207 (545)
R = 500 000
- 3806 (19683)
R = 700 000
R = 1 000 000
Applying the net-benefit regression approach
Table 3 presents the results of the overall economic evaluation with a range of arbitrary ceiling ratios (arbitrary but selected around the sample ICER). We use place of residence (Nouna town versus Nouna villages) as examples of covariates and present net-benefits estimates for different values of the ceiling ratio (Ro) including zero. The coefficients were obtained from an Ordinary Least Squares (OLS) regression as explained above and correspond to the increment net-benefit. The aim of this analysis and the results displayed in this table is to demonstrate that the standard descriptive analysis for computing an Incremental Cost-Effectiveness Ratio (ICER) is equivalent to the net-benefit framework, as the linearization of the equation ICER = λ. One can verify that when λ is = 0, the increment net-benefit (nb2 - nb1) = λ*(Average effect1 – Average effect1) – (Average cost2 - Average cost1) = 0*(85 – 71) – (70253 – 9630) = − 60623.
For example, when the ceiling value is 700,000 XOF (equivalent to $1600) for example, the coefficients obtained from the OLS regression for the overall sample, populations in Nouna town, and populations in Nouna villages are 37664 XOF, 77075 XOF, and 20354 XOF respectively. It can be verified manually that these numbers correspond (apart from rounding errors) to what would have been obtained by the equation (nb) = λ* (effect) – (cost), where (nb) is the average net-benefit, 700,000 the ceiling ratio (λ), (effect) is the mean effect and (cost) is the mean cost.
Simple net-benefit regression estimates with different ceiling ratios, Nouna community based health insurance, household survey, 2007, Burkina Faso
N = 1344
F (1, 1344)
Prob > F
Cost effectiveness acceptability curves
The cost-effectiveness acceptability curve graphically represents the levels of certainty around the cost-effectiveness analysis ratio of two interventions by plotting hypothetical estimates of ceiling ratios against the probability that a new intervention (the Nouna CBHI scheme in our case) is preferred over an existing intervention (the status quo). As can be seen on Figure 2 if the decision makers in Nouna are only prepared to pay a ceiling ratio less than 400,000 XOF ($920) the probability that the Nouna CBHI scheme is preferred or deemed cost effective is small less than 30%. However, if decision makers in Nouna are willing to pay a ceiling ratio more than 700,000 XOF ($1600) the probability that the Nouna CBHI scheme is preferred or deemed cost effective increases to above 99% (asymptotic and close to but never reaching 100%).
Cost-effectiveness acceptability curves from the net-benefit regression, Nouna CBHI, household survey, 2007, Burkina Faso
Values of ceiling ratio
Treatment (intervention) coefficients
One sided p-value
Probability of cost-effectiveness (OLS regression)
1 000 000
1 500 000
2 000 000
2 500 000
Cost-effectiveness acceptability curves, net-benefit OLS regression, Nouna CBHI (Place covariate)
Values of ceiling ratio
Treatment (intervention) coefficients
One sided p-value
Probability of cost-effectiveness
Treatment (intervention) coefficients
One sided p-value
Probability of cost-effectiveness
1 000 000
1 500 000
2 000 000
2 500 000
Simple net-benefit regression estimates with different ceiling ratios, and covariates adjusted net-benefit regression estimates, Nouna community based health insurance, household survey, 2007, Burkina Faso
N = 1344
Place of residence
F (4 1343)
Prob > F
With the traditional ICER, we concluded that approximately $1000 was the incremental in cost for achieving one additional increase in utilization of health services in the intervention area. However, there was no context specific threshold (ceiling ratio) to indicate if the $1,000 was a good value for money for achieving one households’ extra utilization of health services. This estimate provides no clue for what to do, how to do it or where to do it if one is concerned with equity issues in universal health coverage. What could be more helpful is to have some insights in how the cost-effectiveness vary by some equity determinants (for example covering Nouna villages or Nouna town, or design the intervention by groups of households’ assets ownership) which will be the basis for policy making. By applying the net-benefit framework, we were able to conclude that the probability (adjusting for place) of the Nouna community based health insurance to achieve one extra utilization of health services when the ceiling ratio is approximately $1,000 is barely 30% for Nouna villages whilst the corresponding probability for households living in Nouna town is over 90%. This piece of information has important implication for policy making if decision makers are concerned with achieving high probability of cost-effectiveness of the intervention in the poorest population (Nouna villages) and would not have been possible just by using stratified analysis on the traditional ICER approach. As pointed by Hosh JS  the existence of important sub-groups affects how the cost-effectiveness varies at the margin and need to be accounted for when analyzing and interpreting cost-effectiveness results. This is the case with universal health coverage.
Identifying what intervention is cost-effective compared to an alternative is an important piece of information, and we have now seen that even if one does not know the true value of the ceiling ratio, using a cost-effectiveness acceptability curve (Figure 2) it is possible to show the level of uncertainty surrounding the estimated ICER. Computing the ICER point estimate (assuming we know the context specific ceiling ratio to rule if a good value for money) does not provide clues about what policy makers could do, how to do it or where to do it when there are important subgroups. Although it is possible to use modeling such as bootstrap method with appropriate weighting  to assess how the covariates affect the cost-effectiveness at the margin, modeling will rely on data from secondary sources, instead of actual household level effect and cost data.
If the decision makers are concerned with meeting the needs of the poorest and equity [9, 17, 18] we argue we need a joint probability distribution of context specific data on cost and effect and the intervention with important determinants of poverty so that policy making be based on the most important determinants (adjusting to known covariates). For example, when examining the net-benefit regression results (Table 7) it can be noted that only a small fraction (3.4%) of the variability in cost-effectiveness can be explained by covariates used in the analysis. This is very small and points to the existence of unknown important variables not captured in our analysis or the low content validity of our constructed metrics. In any case, this type of information is very useful for appraising the likelihood of covariates to affect the cost-effectiveness of the intervention. Although decision making does not revolve around cost-effectiveness analysis results, this type of information may be very useful for policy making and is not possible to have with the more traditional ICER approach. This is the main potential of the net-benefit framework when applying to public health interventions such as universal health coverage.
Lastly, while the main concern of this paper is with methodological challenges, the high values of the ceiling ratios that we found for the CHBI scheme to be cost-effective are of interest. Together with other evidence, such as the lack of any significant reduction in mortality rates, the low enrolment rates [9, 10], and the high costs of the scheme  cast great doubt upon its cost-effectiveness. Our findings thus challenge the currently fashionable assumption that community based health insurance schemes are a very promising way to extend access to health care in low and middle income countries.
From a methodological point of view, the net –benefit approach opens up the possibility of a marriage of epidemiological, demographic and econometric analytical frameworks in appraising the monetary values of public health interventions. Its application requires the availability of person-level or household-level effect and cost data collection, which in turn will require change to current well established tools and methods of national or sub national household surveys. Although this work was focused on using some of the properties of regression techniques for utilization of health services, there is a possibility, with increasing emphasis on valuing health outcomes in the developing world, to actually use the full potential of regression frameworks and predict for example the net-gains of interventions in saving the lives of women and newborns.. Given this paper is mainly to demonstrate feasibility and applicability of the net-benefit approach and data implications we did not extend on other model diagnosis tests such as normality of residuals.
There are some challenges in interpreting the traditional ICER results. Regardless of the accuracy of the costing methodology, there remain issues about the level of certainty of the computed estimate. The decision rule of the ICER estimate requires knowledge as to whether the estimate is below or above an externally set value which is the maximum decision makers will be willing to pay for an extra unit of health gain. This value is unknown in most cases but in these circumstances, the net-benefit approach has proven feasible and more insightful in assessing cost-effectiveness of a public health intervention aiming at universal health coverage. The net-benefit approach has the relative advantages of better presentation and interpretation of cost-effectiveness analysis results. The most important advantage in our view is the possible information on the marginal cost-effectiveness of important covariates particularly when decision makers are concerned with important determinants or equity as is often the case in public health interventions. However, its applicability requires appropriate data sets (household-level effect and cost data) which will require that we revise the traditional methods and tools of household surveys to ensure concurrent collection of household-level effect and cost data.
Sennen Hounton is a medical epidemiologist with expertise in maternal and newborn health, health systems and economic evaluation. Sennen Hounton is has an MD (Benin), MPH in Epidemiology (University of Oklahoma, USA) and a PhD in Public Health from University of Aberdeen, (Scotland, UK). He was a Senior Research Fellow with Immpact (Initiative for Maternal Mortality Program Assessment). He is currently Maternal Health Technical Adviser at the United Nations Population Fund (New–York), and served as Scientific and Technical Advisor on the WHO Alliance for Health Policy and System Research Scientific and Technical Advisory Committee.
David Newlands is a Senior Lecturer in Economics, previously Team Leader of the Economic Outcomes of an international research programme – Immpact. He is currently a Senior Lecturer at the Business School of University of Aberdeen, Scotland, UK.
This work was undertaken as part of Dr Sennen Hounton’s doctoral programme, funded by the international research programme – Immpact, University of Aberdeen. A special thanks to Nicolas Meda, Peter Byass, and Bocar Kouyate. The funders have no responsibility for the information provided or views expressed in this paper. The views expressed herein are solely those of the authors.
- Drummond MF, Sclpher MJ, Torrance GW, O’Brien BJ, Stoddart GL: Methods for the economic evaluation of health care programmes. Oxford University Press, New-York; 2005.Google Scholar
- Hoch JS, Briggs AH, William AR: Something old, something new, something borrowed, something blue: a framework for the marriage of health econometrics and cost-effectiveness analysis. Health Econ 2002, 11: 415–430. 10.1002/hec.678PubMedView ArticleGoogle Scholar
- Hoch JS, Rockx MA, Krahn AD: Using net-benefit regression framework to construct cost-effectiveness acceptability curves: an example using data from a trial of external loop recorders versus Holter monitoring for ambulatory monitoring of “community acquired” syncope. BMC Health Serv Res 2006, 6: 68. 10.1186/1472-6963-6-68PubMed CentralPubMedView ArticleGoogle Scholar
- Hoch JS, Dewa CS: Lessons from trial-based cost-effectiveness analysis of mental health interventions. Why uncertainty about outcomes, estimates and willingness-to-pay matters. Pharmacoecnomics 2007,25(10):807–816. 10.2165/00019053-200725100-00001View ArticleGoogle Scholar
- Hoch JS, Blume JD: Measuring and illustrating statistical evidence in a cost-effectiveness analysis. J Health Econ 2008, 27: 476–495. 10.1016/j.jhealeco.2007.07.002PubMedView ArticleGoogle Scholar
- Claxton K: The irrelevance of inference: a decision making framework to the stochastic evaluation of health care technologies. J Health Econom 1999, 18: 341–364. 10.1016/S0167-6296(98)00039-3View ArticleGoogle Scholar
- Briggs A, Fenn P: Confidence intervals or surfaces? Uncertainty of the cost-effectiveness plane. Health Econ 1998, 7: 723–740. 10.1002/(SICI)1099-1050(199812)7:8<723::AID-HEC392>3.0.CO;2-OPubMedView ArticleGoogle Scholar
- Briggs AH, Gray AM: Handling uncertainty when performing economic evaluation of healthcare interventions. Health Technol Assess 1999,3(2):1–134.PubMedGoogle Scholar
- Dong H, De Allegri M, Gnawali D, Souares A, Sauerborn R: Drop-out analysis of community-based health insurance membership at Nouna, Burkina Faso. Health Policy 2009,92(2–3):174–179.PubMedView ArticleGoogle Scholar
- Dong H, Gbangou A, De Allegri M, Pokhrel S, Sauerborn R: The differences in characteristics between health-care users and non-users: implication for introducing community-based health insurance in Burkina Faso. Eur J Health Econ 2008,9(1):41–50. 10.1007/s10198-006-0031-4PubMedView ArticleGoogle Scholar
- Dong H, Kouyate B, Cairns J, Mugisha F, Sauerborn R: Willingness-to-pay for community-based insurance in Burkina Faso. Health Econ 2003,12(10):849–862. 10.1002/hec.771PubMedView ArticleGoogle Scholar
- Dong H, Kouyate B, Cairns J, Sauerborn R: Differential willingness of household heads to pay community-based health insurance premia for themselves and other household members. Health Policy Plan 2004,19(2):120–126. 10.1093/heapol/czh014PubMedView ArticleGoogle Scholar
- De Allegri M, Sauerborn R: Community based health insurance in developing countries. BMJ 2007,334(7607):1282–1283. 10.1136/bmj.39240.632963.80PubMedView ArticleGoogle Scholar
- Lothgren M, Zethraeus N: Definitions, interpretation and calculation of cost effectiveness acceptability curves. Health Econ 2000, 9: 623–630. 10.1002/1099-1050(200010)9:7<623::AID-HEC539>3.0.CO;2-VPubMedView ArticleGoogle Scholar
- Van Hout VA, Al MJ, Gordon GS, Rutten FF: Costs, effects and C/E ratios alongside a clinical trial. Health Econ 1994, 3: 309–319. 10.1002/hec.4730030505PubMedView ArticleGoogle Scholar
- Schroeder E, Petrou S, Patel N, Hollowell J, Puddicombe D, Redshaw M, Brocklehurst P, Birthplace in England Collaborative Group: Cost effectiveness of alternative planned places of birth in woman at low risk of complications: evidence from the Birthplace in England national prospective cohort study. BMJ 2012, 344: e2292. 10.1136/bmj.e2292PubMed CentralPubMedView ArticleGoogle Scholar
- Ranson MK, Sinha T, Chatterjee M, Gandhi F, Jayswal R, Patel F, Morris SS, Mills AJ: Equitable utilisation of Indian community based health insurance schemes among its rural membership: cluster randomised controlled trial. BMJ 2007,334(7607):1309. 10.1136/bmj.39192.719583.AEPubMed CentralPubMedView ArticleGoogle Scholar
- Chowdhury ME, Ronsmans C, Killewo J, Anwar I, Gausia K, Das-Gupta S, Blum LS, Dieltiens G, Marshall T, Saha S, Borghi J: Equity in use of home-based or facility-based skilled obstetric care in rural Bangladesh: an observational study. Lancet 2006,367(9507):327–332. 10.1016/S0140-6736(06)68070-7PubMedView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.