This study shows that, given the parameters used, increasing case finding by 20% would lead to nearly 10 times the numbers of deaths saved by the current direct observation of therapy in India. This is due to a greater RRR, applied to a higher baseline case-fatality, hence producing a larger absolute benefit. The analysis clearly shows that, based on the assumptions made, improving case-finding for TB will save far more lives than maximizing the use of the directly observed component of DOTS among currently identified cases in India. This has obvious implications for health decision-makers. Although the DOTS program does encourage better case detection, it does not include active or enhanced case-finding [5, 6, 20]. Our findings concur with those who feel that case finding should receive more attention [7, 21, 22], and that a careful approach to examining the benefits of different aspects of DOTS programmes should be encouraged.
There are several different strategies for increasing case-finding, and the choice should depend on local factors such as disease prevalence, adequacy of training of healthcare workers and willingness of the infected population to seek care for symptoms. The benefits of active, passive or enhanced case-finding among general, high-risk, and symptomatic populations have been reviewed in depth by Golub et al.
The calculations made in this study depend on the accuracy of the data used. Each estimate is subject to potential error, and ideally, each population should obtain its own data in order to produce accurate estimates of population impact of risks and interventions. The literature-based estimates of RRR are also open to question. Our use of the baseline risk of death from treated and untreated TB uses old trial data, although the treated risk is consistent with recently published data of death rates of 4.4% among DOTS-registered cases in countries with high TB burden. We applied the cure rate of the systematic review of DOTS therapy to the case fatality, and this may be open to question. An updated Cochrane Review was published in 2006, which came to the same conclusion about a non-significant 6% difference in the outcome cured or completed treatment. There were insufficient numbers to allow analysis of case fatality, although four of the included trials did include mortality as an outcome measure. The results of the calculations depend on the various assumptions made. For example, changing the estimate of TB treatment completion would have changed the number of deaths prevented, but is unlikely to have changed the ranking of benefit between increasing case-finding and direct observation of treatment. Thus it is important to obtain relevant local data, apply the approach to different populations and population subgroups, and to test the robustness of the estimates to varying the assumptions. For example, in a population where case finding is already extensive, costs per case detected are likely to be higher, and thus maximizing the use of directly observed therapy could have the greater impact.
The confidence intervals were made taking into account the potential variability of the estimates used; however, changing the estimates used for each of the components of the PIMs themselves has additional potential to influence the results. This can be explored by recalculating the measures using different estimates as appropriate to particular local settings. To assist potential users, we have developed a public access website that automatically calculates both the PIMs and their confidence intervals http://www.phsim.man.ac.uk/.
We did not include the impact of transmission dynamics on the rest of the population in our calculations, as this would add considerably to the complexity of the approach. In terms of the comparison between direct observation and improved case-finding, the lack of attention to transmission dynamics will underestimate the relative benefit of improved case-finding, as the resulting early intervention is likely to be more effective than the later attempt to improve therapy adherence. The decision to restrict the analysis to smear-positive cases was based on the better accuracy of this measure of disease burden, although it will underestimate the benefit of treatment on extrapulmonary TB. We do not wish to claim false accuracy for the results presented; however, they do use the best available data, are easy to compute and produce results that are easy to understand. The level of accuracy should be adequate for most policy decisions, and is preferable to making decisions in the absence of estimates of population benefit.
If PIMs prove useful where relevant data exist, then this could stimulate better collection and use of health data in situations where there are currently too few data to enable reasonable calculations. Similarly, the use of PIMs with confidence intervals enables policy-makers to be explicit about the uncertainties they face.
The measures can be used to compare between populations, in which case standardization for age, sex, ethnicity and socioeconomic status may be required. The measures can also be used to compare between segments of a population, in which case the potential for an intervention to reduce health inequalities and increase equity within a population can be explored.
For ease of presentation, the population denominator we used was 100 000 adults; however, one of the main benefits of these measures is the ability to relate to local context, hence policy-makers can make the calculations for their own population denominator.
The addition of information on costs to the population impact measures will also be important for policy makers, and our estimated costs for the programme will allow costs and their consequences to be calculated. PIMs differ from measures used in cost-effectiveness analyses by not including life expectancy or the utility or valuation of the health outcome. They thus produce outcomes expressed as events rather than generic outcomes such as QALY or DALY gained. We have previously suggested that the valuation of the events prevented should be performed by the policy-maker in relation to the costs of the intervention, once the measures have been produced . Although our cost estimates are prone to error, they were derived using previously published costs applied to this simple model of health gain. They show that the programme costs for the direct observation of therapy and the detection of new cases are similar, although the health gain is much greater for the detection of new cases. In fact, our estimate of the health gain from direct observation may be an overestimate, as the relative risk quoted was not significantly different from no effect.
The data shown on TB control demonstrates the potential benefit to policy-makers from the use of PIMs. There is some evidence that the way in which health benefits are framed impacts on the policy decisions reached, although there is debate about the consistency of such an effect. We previously showed that although clinicians were more influenced by benefit expressed as RRR than as PIMs, public health professionals were more likely to prioritise interventions based on the use of PIMs than on the basis of more complex demonstrations of health gain . The evidence base on what are the most effective methods of presenting health gains from interventions in order to assist in policy-making is, however, weak. Health policy requires more than merely demonstration of health gain [30–32], as the complexity of policy-making includes social and political drivers of decision-making, as well as the need to take into account issues such equity, total budget impact, total morbidity and disease severity.
Our method is considerably simpler than other modelling exercises that have also examined the potential benefits of case finding[7, 8], and may compensate for this simplicity by the increased ease of when making the calculations and understanding the results. Both modelling and the calculation of PIMs have similar reliance on data availability and accuracy. Our method is not intended to replace more complex analyses of TB control, nor is it intended to reflect adversely on the DOTS programme, which has many components, of which direct observation is only one. It is intended to show how to use local data to provide simple demonstrations of health benefit, which can then be used in policy-making. We believe that it is worth testing the hypothesis that simple demonstration of the population health-gain consequences of interventions will lead to the introduction of health policies that can provide appropriate priorities to improve health in developing countries. This could be of marked benefit in TB, a major threat to global health.