NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Collaborating Centre for Chronic Conditions (UK). Tuberculosis: Clinical Diagnosis and Management of Tuberculosis, and Measures for Its Prevention and Control. London: Royal College of Physicians (UK); 2006. (NICE Clinical Guidelines, No. 33.)

  • This publication is provided for historical reference only and the information may be out of date.

This publication is provided for historical reference only and the information may be out of date.

Cover of Tuberculosis

Tuberculosis: Clinical Diagnosis and Management of Tuberculosis, and Measures for Its Prevention and Control.

Show details

Appendix HHealth Economic Models

Economic analysis of diagnostic tests for latent infection

INTRODUCTION

This analysis addresses the question of whether the newer interferon gamma tests (IGT) based on ESAT-6 and CFP antigens offer a more cost-effective means of identifying patients with suspected latent infection to receive treatment for latent TB infection compared with conventional tuberculin skin tests (TST) based on PPD. We do not compare different types of skin tests or different types of interferon gamma tests. Thus, throughout this paper the term ‘TST’ refers to either Heaf or Mantoux skin tests using PPD and standard rules for interpretation of the result in the presence of absence of prior BCG, and ‘IGT’ refers to either T-Spot or QuantiFERON-TB Gold commercial immunological tests, which use a combination of ESAT-6 and CFP-10 antigens.

A decision model is used to compare the expected costs (£) and health effects (QALYs) of four strategies of testing in the context of a contact tracing programme in England and Wales. The strategies compared are: a) TST; b) IGT; c) TST followed by IGT for patients with a positive TST; and d) no test (inform and advise only). It is assumed that treatment follows current policy: with appropriate therapy for people diagnosed with active TB, treatment for latent TB infection for those testing positive for latent infection, and BCG when appropriate for others.

Various assumptions are made about the epidemiology and likely concordance with testing and treatment programmes. However, these factors will vary with the context of contact tracing. There is also considerable uncertainty over the relative accuracy of the TST and IGT tests, as well as over some of the other model parameters. Sensitivity analysis is used to explore the impact of such variations and uncertainties.

The model used is a decision tree, which does not account for the dynamics of disease transmission within the population. Instead, for simplicity, we assume that each primary case of active disease is associated with a fixed number of secondary cases. This is probably a reasonable assumption when comparing tests with similar sensitivity, since the absolute difference in false negatives, and hence in opportunities for transmission within the community, will be small. However, estimates of the relative cost-effectiveness of contact tracing per se are less robust and should be treated with caution.

METHODS

Estimation of test accuracy

The effectiveness of the tests largely depends on their accuracy (sensitivity and specificity). Direct comparisons of IGT with TST suggest that the ESAT-6/CFP immunological tests are likely to be more specific than the PPD skin tests due to the confounding of the latter with prior BCG status. The greater correlation of the IGT with exposure also suggests that they are likely to be more specific (and possibly also more sensitive?), than TST. However, as discussed in the clinical review, the absolute sensitivity and specificity of the tests is difficult to assess in the absence of a ‘gold standard’ test for latent infection.

One possible approach would be to treat IGT as the new gold standard. However, unless it really does have 100% sensitivity and specificity, this would artificially inflate its cost-effectiveness. Another possibility would be to use estimates of sensitivity and specificity from studies in selected populations (active disease and very low-risk respectively)20, but it not clear that these results translate to the population with suspected infection in the context of contact tracing.

The approach that we take here is a ‘what if’ analysis, in which we assume values of sensitivity and specificity for IGT (Se and Sp) and then comparative losses of sensitivity and specificity for TST (LossSe and LossSp). The baseline values of Se, Sp, LossSe and LossSp are then varied and the impact on costs and effects is estimated.

Figure 1. Relative accuracy of TST and IGT.

Figure 1Relative accuracy of TST and IGT

Testing Strategies

The decision tree is shown in Appendix 1. We start with a cohort of people suspected of having latent TB infection, as identified from a contact tracing programme. The prevalence of latent infection in this cohort (prev) depends on their characteristics (age, prior BCG, ...) and the circumstances within which they have been identified (degree of exposure to an infectious case, time since exposure, ...).

Four possible testing strategies are compared:

  1. TST
    We assume that a TST is administered to the whole cohort, but that only some patients (pRead) return to have the result read. This parameter is intended to reflect the final proportion with a TST result, including those who have to have a repeat skin test administered because, for whatever reason, the result of their first test is not read within the necessary time. To estimate the average cost of administering skin tests we assume that some proportion (pRepeat) require a second test. Patients who do not end up with a TST result are assumed to have the same prevalence of latent infection and of active disease as those who do.
    Patients with a TST result have a probability p1 of a positive finding, where p1 is a function of the prevalence of active or latent infection in the cohort being tested (prev), and the sensitivity and specificity of TST:
    p1=((prev*(Se-LossSe))+((1-prev)*(1-(Sp-LossSp))))
    The proportion of those who test positive who really have a TB infection is called the ‘Positive Predictive Value’, p2:
    p2=(prev*(Se-LossSe))/((prev*(Se-LossSe))+((1-prev)*(1-(Sp-LossSp))))
    Similarly, the proportion of those who test negative who do not have a TB infection is the ‘Negative Predictive Value’, p3:
    p3=((1-prev)*(Sp-LossSp))/((prev*(1-(Se-LossSe)))+((1-prev)*(Sp-LossSp)))
    There are five possible outcomes from this strategy: 1) true positive; 2) false positive; 3) false negative; 4) true negative and 5) no test. The costs and consequences of these outcomes are described below.
  2. IGT
    With IGT the sequence of possible events and outcomes is similar to scenario a, except that it is assumed that fewer patients will be lost to follow-up, since there is no need for them to return to have the result of the test read. The sensitivity and specificity of IGT are also assumed to be higher than those of TST. The formulae for the probability of a positive test (p4), the Positive Predictive Value (p5) and the Negative Predictive Value (p6) of IGT are:
    p4=((prev*Se)+((1-prev)*(1-Sp)))p5=(prev*Se)/((prev*Se)+((1-prev)*(1-Sp)))p6=((1-prev)*Sp((prev*(1-Se))+((1-prev)*SP))
  3. IGT for TST positive patients
    Here we assume that all patients are given a TST, followed by an IGT only for those who are TST positive. The proportion of positive results and the Negative Predictive Value for the skin test are p1 and p3, defined as above. The prevalence of infection in the subgroup of patients with a positive TST result is p2, the positive predictive value of TST. We assume that within this group the sensitivity and specificity of IGT is the same as in the wider population with suspected infection (Se and Sp). The chance of a positive result and the positive and negative predictive values of IGT are thus:
    p7=((p2*Se)+((1-p2)*(1-Sp)))p8=(p2*Se)/((p2*Se)+((1-p2)*(1-Sp)))p9=((1-p2)*Sp)((p2*(1-Se))+((1-p2)*Sp))
  4. No test (inform and advise only)
    Finally, for comparison, we include the possibility of no testing.

Treatment outcomes

There are five possible outcomes from the above strategies: 1) true positive; 2) false positive; 3) false negative; 4) true negative and 5) no test.

Those patients who test positive, including both true and false positives, are assessed for active TB. We assume that this incurs a cost of one visit to a respiratory medicine clinic, and that this is 100% accurate at detecting active disease, which occurs in a proportion (pTB) of infected patients. Patients diagnosed with active TB are treated according to the guideline recommendations, using the same resource use and cost assumptions as in the schools BCG model. Early diagnosis of active disease is assumed to reduce the QALY loss for the individual and also the number of secondary cases.

Patients with a positive test result but no active disease are offered treatment for latent TB infection, although only some of them (pPro) actually commence treatment. Without prophylaxis, patients with a latent infection have a greater chance of later contracting active disease, compared with non-infected members of the cohort (relative risk rrLatent). With prophylaxis, this chance is reduced (relative risk rrPro). We assume that patients without current latent infection have a baseline risk r15 of contracting TB over the fifteen year time horizon of the model, and that they derive no benefit from prophylaxis.

We assume that patients with a negative test result have no further assessment, but are offered BCG vaccination if appropriate (ie if they are aged 35 or younger and there is no evidence of prior vaccination). The proportion of TST- patients given BCG is pBCG. BCG vaccination is assumed to reduce the risk of later TB disease for uninfected individuals (relative risk rrBCG), but offers no protection for those who are already infected.

Economic analysis

The analysis is conducted from an NHS perspective. A time horizon of 15 years is used for the analysis, chosen to reflect the duration of BCG protection. The costs and outcomes are discounted (where appropriate) using an annual rate of 3.5%. We assume an average time delay of lagLatent years before people with latent infection who go on to develop active disease do so. Similarly, where TB occurs in later life for people without current latent infection or active disease, this is assumed to occur after an average time delay of lagNoInf years.

We assume that each case of active disease is associated with nSec secondary cases, which occur on average lagSec years after the index case. However, the number of secondary cases is assumed to be reduced (by a proportion pSec) when the index case is detected early through contact tracing. Sensitivity analysis is used to investigate variation between contact tracing programmes and uncertainty over modelling parameters. In the base case analysis, the best point estimate for each parameter was used. Each parameter was then varied between a lower and upper limit, chosen to reflect the extent of uncertainty or possible variation.

Input parameters

The parameters required to evaluate the model were estimated from various sources. Where possible, parameter values are based on best available evidence from the literature. However, figures were not available for all the necessary parameters. In these cases subjective judgements were made by the guideline economist (and checked by the GDG). The base case, range for sensitivity analysis, and source for each input parameter are listed in Annex 2 below.

ANNEX 2. Input parameters.

ANNEX 2

Input parameters.

RESULTS

Basecase results

The basecase analysis is shown in Table 1. This shows that the two-stage strategy (TST/IGT) is within the range usually considered ‘cost-effective’, since the Incremental Cost-Effectiveness Ratio (ICER) is around £20–30,000 per QALY. Compared with this, IGT is not cost-effective (over £1m per QALY gained). TST is both less effective and more expensive than all of the other options (it is ‘dominated’). This basecase analysis is illustrated in Figure 2.

Table 1. Base case results.

Table 1

Base case results.

Figure 2. Cost-effectiveness graph (basecase analysis).

Figure 2

Cost-effectiveness graph (basecase analysis).

Variation in results with prevalence of infection

These results are highly dependent on the context of the contact tracing scheme – with a higher-risk cohort of contacts, the expected benefits of early diagnosis of active cases, prophylactic treatment of latent infection, and vaccination will be greater. This is illustrated in Table 2, which shows the estimated costs and effects of the four strategies as we vary the prevalence of infection within the cohort (holding all other parameters constant at their basecase level). If we assume a cost-effectiveness threshold of £30,000 per QALY (at the upper limit of what is considered appropriate for the NHS), then none of the testing strategies is cost-effective below a prevalence of TB infection of about 8%. At intermediate levels of prevalence (between about 8% and 40%), the two-stage TST/IGT strategy is cost-effective. Above 40% IGT on its own is the most cost-effective option.

Table 2. Results by prevalence of infection (active and latent) in the cohort.

Table 2

Results by prevalence of infection (active and latent) in the cohort.

Variation in results with relative specificity of tests

The results also depend on the assumptions that are made about the relative accuracy of the two types of test. Table 3 shows how the results vary with the difference in specificity between the IGT and TST (from no difference to a difference of 40 percentage points) and with the prevalence of infection in the cohort. All of the other input parameters are held at their basecase levels for this analysis, including the assumed sensitivities of the two tests (90% for TST and IGT). The shaded cells show the most cost-effective options, using a threshold of £20–30K:

Table 3. ICER by prevalence of infection and relative specificity of tests.

Table 3

ICER by prevalence of infection and relative specificity of tests.

  • It can be seen that at 4% prevalence none of the testing options is cost-effective.
  • At 10% prevalence, TST/IGT falls below the upper £30K limit, provided that the specificity of TST is no more than about 10 percentage points lower than the specificity of IGT.
  • At 20% prevalence, TST/IGT falls below the £30K limit at all levels of specificity tested.
  • At 40% prevalence, IGT falls under the £30K threshold if the specificity of TST is 20 percentage points or more less than that for IGT.
  • And at 60% prevalence, IGT falls under the £30K threshold at all levels of specificity tested.

Variation in results with relative sensitivity of tests

Similarly, we show how the cost-effectiveness results change with the assumptions about the relative sensitivity of the TST and IGT (Table 4). Again, we hold all other parameters fixed at their base case levels (including 90% and 80% specificity for IGT and TST respectively).

Table 4. ICER by prevalence of infection and relative sensitivity of tests.

Table 4

ICER by prevalence of infection and relative sensitivity of tests.

  • At 4% prevalence, no testing is cost-effective.
  • At 10% prevalence, TST/IGT falls under the £30K threshold if the sensitivity of TST is no more than 20 percentage points different than that of IGT.
  • At 20% prevalence, TST/IGT falls below £30K at all levels of sensitivity tested.
  • At 40% prevalence, TST/IGT is the optimal strategy, unless the sensitivity of TST is 20 percentage points or more worse than that of IGT, in which case IGT falls under the £30K threshold.
  • And at 60% prevalence, TST/IGT is the optimal strategy if the sensitivity of TST is 10 percentage points better than that of IGT. Otherwise, IGT is the optimal strategy.

Uncertainty over other parameters

As might be expected, given how close the basecase model is to the borderline of cost-effectiveness, the results are also sensitive to most of the other input parameters. The sensitivity of the results is illustrated in Table 5. This shows the absolute difference in the expected net benefit of the optimal strategy (calculated using a cost-effectiveness threshold of £30,000) as each parameter is varied from its lower to its upper limit.

Table 5. One-way sensitivity analysis.

Table 5

One-way sensitivity analysis.

It can be seen that the model is particularly sensitive to some other parameters that define the current and future risk of TB within the cohort (the risk of TB in the general population and relative risk of TB in currently uninfected contacts, and the proportion of currently infected patients with active disease) and also to the rate of transmission (mean number of secondary cases per primary case). The results also change greatly with the expected QALY loss and cost of treating each case of active TB.

CONCLUSIONS

This analysis has demonstrated the very great degree of uncertainty over the cost-effectiveness of contact tracing strategies. The results were highly sensitive to assumptions about the relative accuracy of the two types of test, the risk of current and future TB in the cohort, the level of transmission to the wider population, and also to the expected net benefit of avoiding each active case of TB.

Under the basecase assumptions, the two-stage TST/IGT strategy appears to be cost-effective. This result depends greatly on the circumstances of the contact tracing exercise. Where the target group is at relatively low risk, it is possible that none of the strategies tested is cost-effective. Alternatively, for a high-risk cohort, one step IGT testing may be the most cost-effective strategy, particularly if the sensitivity and/or specificity of TST is a lot worse than that of IGT.

Flowchart Icon

ANNEX 1. Decision tree (PDF, 31K)

Economic analysis of school-based BCG

INTRODUCTION

This paper presents the final version of the economic analysis of the school BCG programme in England and Wales. The model has been revised in line with discussion at the SG11 preventive sub-group meeting.

For this analysis, we define a ‘high-risk’ group consisting of children who should have already been offered BCG before the school programme. This might include: 1) children eligible for selective neonatal BCG programmes (from high-incidence ethnic groups); 2) children eligible under new entrant schemes before the age of 10 (from high-incidence countries); and 3) other children covered by universal neonatal programmes in high-incidence areas. We also define a ‘low-risk’ group as the remainder of the 10–14 year old cohort: those not eligible for any prior vaccination programme. The school BCG programme is potentially beneficial for low-risk children, but also as a catch-up for high-risk children who for whatever reason have not been previously vaccinated. The model described below allows us to estimate the costs and effects of school BCG for both of these groups. It does not provide any information about the cost-effectiveness of neonatal BCG, or about the value of BCG for arrivals from high-risk countries.

The model is a simple decision tree that estimates the number of primary cases for a cohort of 10–14 year olds, the consequent number of secondary cases in the population, and the associated costs and health outcomes, with and without a school BCG programme. Estimates of the impact of school BCG for white 10–14 year old children have been provided by Saeed et al (2002), updating the work of Sutherland and Springett (1989). These estimates are used in our model to estimate the benefits of vaccination for the low-risk group. The benefits for unvaccinated high-risk children are then estimated by varying the proportion of high-risk children in the cohort, the proportion of high-risk children previously vaccinated, and the relative risks for the high-risk group.

It is also important to note that this methodology can only give approximate results for an infectious disease such as TB. A population dynamic model would be expected to provide more reliable results. However, we do not have the resources or time required to build such a model for the NICE guideline. The present analysis will, however, provide a broad indication of the range of cost-effectiveness for different areas. It will also suggest how sensitive the results are to various parameters, which should help to focus further work.

METHODS

Structure of the model

The decision tree is shown below (Figure 3). A cohort of 10 to 14 year old children enters the tree from the left. Some of the high-risk group will have already been vaccinated. It is assumed that these children are easily identified and incur no extra costs (or benefits) for the schools programme. Some proportion of the low-risk and unvaccinated high-risk children participate in the school programme - attending for the initial tuberculin skin test (TST) and then returning to have the result read and BCG if indicated. Those with a negative TST reaction are given the BCG vaccination, and consequently have a somewhat lower chance of contracting TB in future years than if they had not been vaccinated. Children with an intermediate TST reaction are assumed to be immune, and have a lower chance of subsequently contracting TB. Finally, those with a strongly positive skin reaction are referred to an outpatient clinic for evaluation, and may be treated for active TB, given treatment for latent TB infection, or discharged with no further action. The incidence of TB for unvaccinated TST- high-risk children is estimated by multiplying the incidence for the low-risk group (rL) by a relative risk (rrHigh).

Flowchart Icon

Figure 3. Decision tree (PDF, 21K)

The model is evaluated by attaching probabilities to each branch of the tree and values to the end nodes on the right. In Figure 1, the outcomes shown (0 or 1) indicate presence or absence of a primary case of active disease. The tree can thus be used to estimate the expected (mean) number of primary cases with the school BCG programme. Similarly, the model is adapted to estimate the total number of TB cases (primary and secondary), the resultant health outcomes (loss of quality adjusted life years, QALYs), and health care costs (for both the BCG programme and future health care). The formulae for the QALY and calculations are shown in Appendix 1 and those for the cost outcomes in Appendix 2.

Input parameters

The probability, QALY and cost parameters used in the model are summarised in Table 1, Table 2 and Table 3 respectively. These parameters have been estimated from a variety of sources. Where possible, the estimates are based on recent, UK-relevant data. However, where no such information could be found, the values and ranges are based on subjective estimates by the guideline economist and members of the GDG. See Appendices 3–5, for a full list of sources and assumptions.

Methods of economic evaluation

The evaluation followed the methods of the NICE ‘reference case’. Amongst other things, this means that an NHS perspective has been adopted. Costs are estimated for the organisation and delivery of the vaccination programme, as well as savings from reduced treatment of TB disease. However costs or savings to individuals or other organisations are not included.

The time horizon for the analysis has been chosen to reflect the expected duration of benefit from BCG vaccination in the school programme. Analysis of the MRC trial suggests a maximum protection of fifteen years, whereas meta-analysis of all trial data only supports a more limited estimate of ten years. The economic analysis has been conducted for both ten year and fifteen year protection (with estimated of cases prevented and associated costs and QALYs from age 15 to 24, and from age 15 to 29).

Both costs and non-monetary health effects were discounted at an annual rate of 3.5%, which reflects the current UK Treasury recommendations.

There is considerable uncertainty over the values of some of the input parameters. Sensitivity analysis was used to estimate the impact of this uncertainty. Firstly, a simple one-way sensitivity analysis was performed for each parameter – varying its value from the lower to the upper limit shown in Tables 13. Secondly, a probabilistic sensitivity analysis was performed, using distributions listed in these tables.

Table 6Programme effectiveness and epidemiology parameters used in the model

ParameterNameBase caseRange for simple sensitivity analysisDistribution for probabilistic sensitivity analysis
DistributionSEParameters
LowerUpperP1P2
Proportion of cohort in low-risk grouppLow85%80%100%beta7.65%18.503.27
Proportion who are TST+ at school BCGpTSTp7.70%7.61%7.78%beta0.04%30,032360,082
Proportion of TST+ at school BCG who are referredpTSTpp0.5%0.0%1.0%beta0.26%3.82760.66
Proportion of referred children treatedpTreat10%5%20%beta5.10%3.4631.12
Proportion of non-treated children given prophylaxispPro20%5%30%beta5.10%12.2949.17
Proportion of eligible population attending school programmepAttend64%60%80%beta8.39%20.9312.00
Proportion of ‘high-risk’ population previously given BCGpPrior64%60%80%beta7.95%23.3512.90
Baseline risk of TB (age 15–24) in TST- low-risk groupr100.03%0.018%0.045%beta0.01%20.0564,083
Baseline risk of TB (age 15–29) in TST- low-risk groupr150.05%0.028%0.072%beta0.01%20.0639,977
Risk of TB with latent infection (untreated)rLatent1.7%1.5%1.8%beta0.07%557.0032,556
Relative risk of TB for high-risk grouprrHigh401070gamma15.3104.532.61
Relative risk of TB for TST- with BCG (10 year protection)rrBCG100.242910.242880.24295gamma0.0000173,39713,984
Relative risk of TB for TST- with BCG (15 year protection)rrBCG150.181570.181560.18158gamma0.0000048,34745,969
Relative risk of TB (age 15–24) if TST+rrPos100.242910.242880.24295gamma0.0000173,39713,984
Relative risk of TB (age 15–29) if TST+rrPos150.181570.181560.18158gamma0.0000048,34745,969
Relative risk of TB for latent cases with prophylaxisrrPro0.400.310.52gamma0.0612.616.53
Relative risk of transmission if index case detected earlyrrEarly0.50.21gamma0.2550.981.96
Mean secondary cases per primary casenSec0.750.371.12gamma0.192.943.92
Mean latent infections treated per primary casenPro0.060.040.08gamma0.010.426.62

Table 7QALY estimates used in the model

ParameterNameBase caseRange for simple sensitivity analysisDistribution for probabilistic sensitivity analysis
DistributionSEParameters
LowerUpperP1P2
QALY loss due to BCG adverse reactionsqBCG0.000050.000000.00046gamma0.00020.0000130.248451
QALY loss for LI detected at school BCGqPro0.050000.000000.10000gamma0.02550.09801.9600
QALY loss for LI detected from index cases (age 15–24)qPro100.040280.000000.08056gamma0.02060.07891.9600
QALY loss for LI detected from index cases (age 15–29)qPro150.036470.000000.07295gamma0.01860.07151.9600
QALY loss for TB (diagnosed early at school BCG)qTBearly0.140320.030870.30239gamma0.08270.23811.6970
QALY loss for TB (age 10–14)qTB0.156990.034540.33830gamma0.09250.26641.6970
QALY loss for primary case (age 15–24)qTB100.166480.036620.35876gamma0.09810.28251.6970
QALY loss for primary case (age 15–29)qTB150.163440.035960.35220gamma0.09630.27741.6970
QALY loss for secondary case (index age 10–14)qSec0.675890.148691.45652gamma0.39831.14701.6970
QALY loss for secondary case (index age 15–24)qSec100.544500.119791.17338gamma0.32090.92401.6970
QALY loss for secondary case (index age 15–29)qSec150.493050.108471.06250gamma0.29050.83671.6970

Table 8Cost estimates used in the model

ParameterNameBase caseRange for simple sensitivity analysisDistribution for probabilistic sensitivity analysis
DistributionSEParameters
LowerUpperP1P2
Cost of school BCG programme (£ per child attending)cTST£8£5£32gamma12.684.5220.597
Cost of vaccination (£ per child vaccinated)cBCG£4£1£11gamma3.405.0271.215
Cost of referral for strongly positive cases (£ per referral)cRefer£252£214£289gamma19.133,30613.145
Cost per primary case: age 10–14 (£ per case)cTB£5,157£1,928£11,598gamma3286.658,0901.569
Cost per primary case: age 15–24 (£ per case)cTB10£4,298£1,607£9,667gamma2739.236,7431.569
Cost per primary case: age 15–29 (£ per case)cTB15£3,891£1,455£8,751gamma2479.926,1051.569
Cost per secondary case (index age 10–14)cSec£5,098£1,906£11,466gamma3249.087,9981.569
Cost per secondary case (index age 15–24)cSec10£4,107£1,535£9,237gamma2617.486,4431.569
Cost per secondary case (index age 15–29)cSec15£3,719£1,390£8,364gamma2370.135,8341.569
Cost per latent case detected through school BCGcPro£494£251£857gamma185.371,3152.664
Cost per treated latent case (index age 15–24)cPro10£398£202£690gamma149.341,0592.664
Cost per treated latent case (index age 15–29)cPro15£360£148£504gamma73.221,7724.919

QALY estimates

A previous version of the model suggested that the results were sensitive to the QALY estimates. Hence, a rather more accurate method has now been used to estimate the QALY loss from cases of TB at different ages. QALY loss due to TB-related mortality was estimated from information about the distribution of incidence by age (Figure 4), the estimated case fatality rate by age (Figure 5), and life expectation by age (Figure 6).

Figure 4. Estimated incidence by age (HPA data, mean cases 1999–2003, curve fitted by cubic interpolation).

Figure 4

Estimated incidence by age (HPA data, mean cases 1999–2003, curve fitted by cubic interpolation).

Figure 5. Estimated case fatality rate by age (HPA data, mean TB deaths and cases 1999–2003, curve fitted by cubic interpolation).

Figure 5

Estimated case fatality rate by age (HPA data, mean TB deaths and cases 1999–2003, curve fitted by cubic interpolation).

Figure 6. Life expectancy by age (GAD life tables, based on 2000–2 data).

Figure 6

Life expectancy by age (GAD life tables, based on 2000–2 data).

The resulting estimates of QALY loss per fatality and per active case of disease are shown below. It can be seen that at younger ages (under 20) the impact of TB-related mortality is less than that of morbidity, whereas for older cases the reverse is true (Figure 7).

Figure 7. Estimated QALY loss per fatality (discounted to age of death).

Figure 7

Estimated QALY loss per fatality (discounted to age of death).

Figure 8. Estimated QALY loss due to mortality per case (discounted to age of death).

Figure 8Estimated QALY loss due to mortality per case (discounted to age of death)

Figure 9. Estimated QALY loss for morbidity and mortality per case (discounted to age of incidence).

Figure 9Estimated QALY loss for morbidity and mortality per case (discounted to age of incidence)

Other modelling assumptions

The model assumes that current levels of BCG coverage and baseline TB risk (for TST negative children who are not vaccinated) continue in those areas where a School BCG programme is retained, regardless of whether the programme is withdrawn from neighbouring areas. In reality, there would be some spill-over of effects.

Patients who survive active TB are assumed not have a recurrence within the time horizon of the analysis.

Once referred to hospital, identification of active TB and latent infection is assumed to be 100% accurate (there are no false positives or false negatives). After effective prophylaxis, the risk of TB is assumed to be the same as for patients with a positive skin reaction.

RESULTS

Basecase analysis

The results of the basecase analysis are shown in Table 4. The schools programme does not appear to be cost-effective for the low-risk group alone – with 0% in the high-risk group, the incremental cost per QALY gained (ICER) is considerably higher than the conventional threshold of around £20,000–£30,000. However, this may change for some areas if we take account of the catch-up benefit for previously unvaccinated high-risk children. Assuming ten-year BCG protection, the schools programme appears to be cost-effective for areas with around 10–14% or more of children in the high-risk group. If we assume fifteen-year BCG protection, school BCG appears cost-effective with around 4–6% or more in the high-risk group. These results are based on the assumption that 64% of high-risk children have been previously vaccinated, and that they have a relative risk of 40 (compared with the low-risk group).

Table 9Cost-effectiveness of School BCG by % of cohort in high-risk group

'High-risk' as % of cohort10 year protection15 year protection
Additional cost (£K)QALYs gainedICER (£/QALY)Additional cost (£K)QALYs gainedICER (£/QALY)
0%£6515£123,557£58810£56,200
2%£6147£83,327£53514£38,312
4%£5789£60,983£48217£27,595
6%£54212£46,767£42921£20,458
8%£50614£36,926£37624£15,364
10%£46916£29,710£32328£11,544
12%£43318£24,192£27031£8,575
14%£39720£19,836£21735£6,200
16%£36122£16,309£16438£4,258
18%£32524£13,396£11142£2,639
20%£28826£10,950£5845£1,270

School BCG appears to be cost-effective for the low-risk population only if their baseline level of risk (from 15–24 or 15–29 years of age) is approximately 0.08-0.1% (Table 5). This compares with current estimates of 0.03% (age 15–24) or 0.05% (age 15–29).

Table 10Cost-effectiveness of School BCG for low-risk group only by baseline risk of TB

Risk of TB over period of protection10 year protection15 year protection
Additional cost (£K)QALYs gainedICER (£/QALY)Additional cost (£K)QALYs gainedICER (£/QALY)
0.03%£6515£123,600---
0.04%£6178£78,700---
0.05%£58410£56,000£58810£56,200
0.06%£55113£42,400£55513£42,600
0.07%£51816£33,200£52316£33,500
0.08%£48518£26,700£49118£26,900
0.09%£45221£21,800£45821£22,000
0.10%£41923£18,000£42623£18,200

One-way Simple Sensitivity Analysis

The results of the one-way sensitivity analysis are shown in the ‘tornado diagrams’ in Figures 10 and 11. These indicate that the cost-effectiveness of the schools’ programme is most sensitive to:

Figure 10. Tornado diagrams (10 year protection).

Figure 10

Tornado diagrams (10 year protection).

Figure 11. Tornado diagrams (15 year protection).

Figure 11

Tornado diagrams (15 year protection).

  • The estimated QALY loss per case of active TB age 15–24 or 29, and for secondary cases resulting from these index cases.
  • The proportion of the population in ‘high-risk’ groups, and the proportion of these who have previously been vaccinated.
  • The mean number of secondary cases per primary case.
  • The baseline level of risk in the low-risk group (TST- unvaccinated), and the relative risk for those in the high-risk group (TST- vaccinated).
  • The mean cost of treating a case of TB age 15–24/29, or secondary cases resulting from such cases.

Probabilistic Sensitivity Analysis

The extent of uncertainty over the mean costs and effects of the school BCG programme is illustrated in Figure 10 for a population with 5% high-risk and assuming only 10 year protection from BCG. In this diagram the dotted line represents the £30,000 per QALY cost-effectiveness threshold – all points to the northwest of this line indicate that the schools programme is cost-effective. The ellipse shows the region of 95% confidence, based on underlying uncertainty about the input parameters for the model. It can be seen that there is a high degree of uncertainty about the cost-effectiveness of the school programme.

This uncertainty is further illustrated by the cost-effectiveness acceptability curves (CEACs) shown below. With no high-risk children in the cohort, there is a high estimated probability that withdrawing the schools programme would be cost-effective: 90% assuming 10 year BCG protection, and 80% with 15 year protection (Figure 11).

With a 5% proportion of high-risk children in the cohort, there is roughly a 75% chance that withdrawing the schools programme would be cost-effective assuming only 10 year protection and a 55% chance assuming 15 year protection (Figure 12)).

Figure 12. Results of probabilistic sensitivity analysis (5% high-risk, 10 year protection).

Figure 12

Results of probabilistic sensitivity analysis (5% high-risk, 10 year protection).

However, with 10% high-risk children, the estimated probability that withdrawing the school programme would be cost-effective is lower: approximately 60% with 10 year protection, and only 40% with 15 year protection (Figure 13).

Figure 13. Cost-effectiveness acceptability curve (0% high-risk).

Figure 13

Cost-effectiveness acceptability curve (0% high-risk).

Figure 14. Cost-effectiveness acceptability curve (5% high-risk).

Figure 14Cost-effectiveness acceptability curve (5% high-risk)

Figure 15. Cost-effectiveness acceptability curve (10% high-risk).

Figure 15Cost-effectiveness acceptability curve (10% high-risk)

CONCLUSIONS

This analysis suggests that the current schools’ BCG programme is only likely to be cost-effective in areas of the country with relatively high proportions of unvaccinated 10–14 year olds in ‘high risk’ groups: approximately 10% or more, assuming that high-risk children have forty times the incidence of low-risk children and that only 64% of high-risk children are vaccinated prior to school BCG. This is most likely to apply to areas such as London and the West Midlands, with relatively large numbers of children from high-incidence ethnic groups or born in high-incidence countries.

This analysis has only considered the costs and consequences of the schools programme. It is therefore unclear whether even in high-incidence areas, the resources used for the schools programme could be better directed towards improving the uptake of neonatal or new entrant schemes, or by introducing universal neonatal programmes.

There is considerable uncertainty over the results of the model due to uncertainty over some of the input parameters for the analysis. In particular, the results are sensitive to the proportion of 10–14 year olds in ‘high-risk’ groups, the proportion of these high-risk children previously vaccinated, the baseline level of risk in the cohort, and the relative risk for the high-risk group. The results are also sensitive to the estimated QALY loss due to TB, and the estimated cost of treating a case of TB. Finally, the results were sensitive to the mean number of secondary cases per primary case. This suggests that more reliable results might be obtained from a population model, reflecting the dynamics of transmission of the disease.

Flowchart Icon

ANNEX 1. Decision tree with QALY outcomes (PDF, 20K)

Flowchart Icon

ANNEX 2. Decision tree with cost outcomes (PDF, 21K)

ANNEX 3Input data and assumptions: programme effectiveness and epidemiology

Base caseLower limitUpper limitSourceComments
Population and baseline risks
 Proportion of cohort in low-risk grouppLow85%70%100%AssumptionRange to reflect possible variation between areas.
 Baseline risk of TB (age 15–24) in TST- low-risk groupr100.031%0.018%0.045%Saeed et al 2002 (table 2a)1 in 3197 - assuming 10 year protection, 2003 cohort. Lower limit shows estimate for 2033 cohort (1:5685)
 Baseline risk of TB (age 15–29) in TST- low-risk groupr150.050%0.028%0.072%Saeed et al 2002 (table 2b)1 in 1994 - assuming 15 year protection, 2003 cohort. Lower limit shows estimate for 2033 cohort (1:3545)
Relative risk of TB for high-risk grouprrHigh401070AssumptionCurrent recommendation for neonatal screening for groups/populations with 40/100,000 (compared with baseline risk for white UK born of around 1 in 100,000). Range to examine variation between areas.
 Mean secondary cases per primary casenSec0.750.371.12Saeed et al 2002 (table 4b)Inferred from estimated numbers of secondary/primary notifications between 2003 to 2023 from stopping BCG at end of 2002 (assuming 10 year BCG protection)
 Mean latent infections treated per primary casenPro0.060.040.08Underwood et al 2003)41 contacts given prophylaxis out of 646 traced.
Coverage of BCG programmes
 Proportion of eligible population attending school programmepAttend64%60%80%DH & NAW 2002–3Estimated from number of skin tests divided by estimated white population in one year cohort (age 10–14)
 Proportion of ‘high-risk’ population previously given BCGpPrior64%60%80%DH & NAW 2002–3Estimated from annual number of vaccinations (age 0–9) divided by estimated high-risk population in one year cohort (assuming (1-pLow) proportion of year group is high risk).
Effectiveness of BCG programmes
Relative risk of TB for TST- with BCG (10 year protection)rrBCG100.242910.242880.24295Saeed et al 2002 (table 2a)1 in 13161/1 in 3197 - assuming 10 year protection, 2003 cohort. Upper limit shows estimate for 2033 cohort.
Relative risk of TB for TST- with BCG (15 year protection)rrBCG150.181570.181560.18158Saeed et al 2002 (table 2b)1 in 10,982/1 in 1994 - assuming 15 year protection, 2003 cohort. Lower limit shows estimate for 2033 cohort.
Relative risk of TB (age 15–24) if TST+rrPos100.242910.242880.24295AssumptionAssumed same as for TST- with BCG: 10 year protection
Relative risk of TB (age 15–29) if TST+rrPos150.181570.181560.18158AssumptionAssumed same as for TST- with BCG: 15 year protection
 Proportion who are TST+ at school BCGpTSTp7.7%7.6%7.8%DH & NAW 2002–3Positive rate for 10–15 age group 2002/3. Range 95% CI.
 Proportion of TST+ at school BCG who are referredpTSTpp0.5%0.0%1.0%Assumption
 Proportion of referred children treatedpTreat10%5%20%Assumption
 Proportion of non-treated children given prophylaxispPro20%5%30%Assumption
Relative risk of transmission if index case detected earlyrrEarly0.50.21.0Assumption
 Risk of TB with latent infection (untreated)rLatent1.7%1.5%1.8%Smieja et al 2004Risk of infection in control groups of included studies from meta analysis (n=33113). Various lengths of follow-up. Range shows 95% CI.
Relative risk of TB for latent cases with prophylaxisrrPro0.400.310.52Smieja et al 2004Relative risk with isoniazid treatment of 6 months or more (95% CI)

ANNEX 4Input data and assumptions: QALY estimates

Base caseLower limitUpper limitSourceComments
Quality of life values
 Population quality of life (0 to 1)QoL0.830.800.90Health Survey for England 1996Mean for 16+ population, men and women.
 QoL loss due to adverse reactions to BCGQoLAR0.100.000.20Assumption
 QoL loss during treatment for latent TB infection for latent infectionQoLPro0.100.000.20"
 QoL loss due to sick time at home with TB (not treated)QoLhome0.100.000.20Schechter, Rose and Fahs 1990Estimates by authors, not from patient survey. Ranges for sensitivity analysis assumed.
 QoL loss due to near-death time in hospital with TBQoLND0.900.801.00""
 QoL loss due to time in hospital with non-fatal TBQoLIP0.500.400.60""
 QoL loss during outpatient treatmentQoLOP0.100.000.20""
Adverse reactions to BCG
 Incidence of adverse reactions to BCGpAR0.3%0.0%1.0%Bannon 1999Reported incidence of suppurative adenitis in older children
 Mean duration of adverse reactions to BCG (years)dAR0.1730.1150.231Marchant 1998Says that localised lesions will heal within 6–12 weeks
 QALY loss due to BCG adverse reactionsqBCG0.000050.000000.00046QoLAR*pAR*dAR
Sickness during treatment for latent TB infection
 Mean duration of treatment for latent TB infection (years)dPro0.50.50.5Guideline recommendations
 QALY loss for LI detected at school BCGqPro0.0500.0000.100QoLPro*dPro
 QALY loss for LI detected from index cases (age 15–24)qPro100.0400.0000.081Discounted to age 15 (for index case) - assumes current distribution of incidence (age 15–24) for index case.
 QALY loss for LI detected from index cases (age 15–29)qPro150.0360.0000.073Discounted to age 15 (for index case) - assumes current distribution of incidence (age 15–29) for index case.
Sickness for active cases
 Time to diagnosis: cases detected by school BCG (years)delayBCG0.0830.0420.125AssumptionOne month
 Time to diagnosis: other cases (years)delay0.2500.1250.375AssumptionThree months
 QALYs lost prior to diagnosis (per case)Qhome0.030.000.08QoLhome*delay
 QALYs gained from early detection at BCG (per case)Qearly0.020.000.05QoLhome*(delay-delayBCG)
 Proportion of cases admitted (all ages)pAdmit53%40%60%HPA & DH dataInpatient episodes/total TB cases
 Mean length of stay for acute TB (days)LoS10.48.0012.00DH Reference CostsHRG D18 non-elective episodes. Close to mean of 10 days for 18 non-MDR cases in White and Moore-Gillon 2000.
 Proportion of time in hospital ‘near death’pND20%10%30%Assumption
 QALYs lost as inpatient for survivorsQIPlive0.0090.0040.014pAdmit*(LoS/365)*(pND*QoLND+( 1-pND)*QoLIP)
 QALYs lost as inpatient for fatalitiesQIPdie0.0260.0260.026(LoS/365)*QoLNDAssumes full time as inpatient is spent ‘near death’.
 Mean duration of outpatient treatment (years)dOP0.50.50.5Guideline recommendationsSix months
 QALYs lost as outpatient (per case)QOP0.0500.0000.100QoLOP*dOP
 QALY loss due to morbidity for TB survivorsQLlive0.08380.00390.1892Qhome+QIPlive+QOP
 QALY loss due to morbidity for TB fatalitiesQLdie0.05070.02570.1007Qhome+QIPdie
Total QALYs lost per active case
 QALY loss for TB (diagnosed early at school BCG)qTBearly0.14030.03090.3024Q-Qearly
 QALY loss for TB (age 10–14)qTB0.15700.03450.3383HPA mortality and incidence (1999– 2003), GAD life expectancy (2000–2)Includes estimated QALY loss due to mortality and morbidity (as estimated above) for primary cases aged 10 to 14.
 QALY loss for primary case (age 15–24)qTB100.16650.03660.3588As above, but for primary cases aged 15–24 (mean weighted by incidence), discounted to age 15.
 QALY loss for primary case (age 15–29)qTB150.16340.03600.3522As above, but for primary cases aged 15–29.
 QALY loss for secondary case (index age 10–14)qSec0.67590.14871.4565Mean QALY loss for secondary cases (mean for all ages weighted by incidence) given that index case occurs between age 10 and 14 and assuming time lag of 1 year for transmission.
 QALY loss for secondary case (index age 15–24)qSec100.54450.11981.1734As above, but for index cases between 15 and 24, discounted to age 15 (for index case).
 QALY loss for secondary case (index age 15–29)qSec150.49300.10851.0625As above, but for index cases aged 15 to 29.

ANNEX 5Input data and assumptions: Cost estimates

Base caseLower limitUpper limitSourceComments
School BCG
 Tuberculin per childucTST£1.22BNF 48, September 20040.1mL Tuberculin + £1 for disposables
 Cost of vaccination (£ per child vaccinated)ucVaccine£3£1£5AssumptionNot publicly available
 Nurse for school BCG session (per hour)uNrs£28PSSRU Unit Costs 2004Assumed equivalent to health visitor cost (including qualification costs and overheads). Uprated for inflation.
 Doctor for school BCG session (per hour)uDr£134"Assumed equivalent to GP per hour of patient contact (including overheads and qualification costs, but excluding costs for other direct care staff). Uprated for inflation.
 School nurse time for skin testing session (hours per child)qNrs10.08Marchant 1998Assumes 3 nurses for 3 hours per school, 117 children per school (estimated from DfES data).
 School nurse time for vaccination session (hours per child)qNrs20.03"1 nurse for 3 hours per school.
 Doctor time for vaccination session (hours per child)qDr0.03"2 doctor for 3 hours per school.
 Clinic visits for treatment of adverse reactionsqOPAR213Assumption by GDG
 GP visits for treatment of adverse reactionsqGPAR102Assumption by GDG
 Cost for first clinic visit (£)ucOP1£252£214£289
 Cost for subsequent clinic visits (£)ucOP2£128£109£146
 Cost per GP visit (£)ucGP£23£10£30
 Cost of school BCG programme (£ per child attending)cTST£8£5£32ucTST+uNrs*qNrs1+uNrs*qNrs2+uDr*q DrCost for both sessions (excluding vaccine). Upper limit from 2003/4 Reference costs
 Cost of vaccination (£ per child vaccinated)cBCG£4£1£11ucTST+pAR*IF(qOPAR>1,ucOP1+(qOP AR-1)*ucOP2,IF(qOPAR=1,ucOP1,0))Includes cost of vaccine and cost of treating adverse reactions (assumes that there are no adverse reactions to the tuberculin test).
 Cost of referral for strongly positive cases (£ per referral)cRefer£252£214£289DH Tariff 2005/6For respiratory medicine speciality. Range for adult and child.
Active cases
Contact tracing
Contact tracing (per contact)ucTrace£317£164£539DH Reference costs 2003/4Assumes one clinic visit per contact (infectious diseases, first visit). Range is interquartile range. Uplifted for inflation.
 Mean number of contacts examined per primary casenContacts6.52.810.2Review of Current Services (see guideline). Underwood et al 2003Midpoint estimate from survey, lower limit from Underwood.
 Cost of contract tracing (£ per primary case)cTrace£2,058£466£5,470ucTrace*nContacts
Inpatient care
 Cost of inpatient episode for acute TB (£ per spell)ucIP£3,457£2,153£6,040DH Tariff 2005/6HRG D18 non-elective (92% of FCEs in 2003/4).
Lower limit for elective cases. Upper limit from White & Moore-Gillon 2000.
 Proportion of cases admitted (all ages)pAdmit53%40%60%HPA & DH Reference CostsInpatient episodes/total TB cases
 Cost of inpatient care (£ per active case)cIP£1,835£861£3,624ucIP*pAdmit
Chemotherapy
 Cost of isoniazid (£ per month)ucIso£12.36£6.18£18.55BNF 48, September 2004Dose: 100, 200, 300mg daily, non-proprietary.
 Cost of rifampicin (£ per month)ucRif£10.76£5.45£21.51"Dose: 150, 300, 600mg daily, non-proprietary.
 Cost of ethambutol (£ per month)ucEth£18.48£12.32£22.89"Dose: 200,300,400 daily, non-proprietary.
 Cost of pyrazinamide (£ per month)ucPyr£6.88£4.58£9.17White and More-Gillon 2000Dose: 1g, 1.5g, 2g daily
 Duration of isoniazid (months)Iso666GDG recommendationRegimen for pulmonary TB (HIV-and non-MDR).
Assumes full concordance with recommended regimen. No DOTS or other increased surveillance assumed.
 Duration of rifampicin (months)Rif666"
 Duration of ethambutol (months)Eth222"
 Duration of pyrazinamide (months)Pyr222"
 Cost of drugs (£ per active case)cdrugs£189£104£304(Iso*ucIso+Rif*ucRif+Pyr*ucPyr+Eth*uc Eth)Close to mean of £150 for 18 non-MDR cases reported in White and Moore-Gillon 2000.
Outpatient care
 Cost of outpatient consultation: first visit (£ per visit)ucOP1£252£214£289DH Tariff 2005/6For respiratory medicine speciality. Range for adult and child.
 Cost of outpatient consultation: follow up visits (£ per visit)ucOP2£128£109£146""
 Cost of TB nurse home visit (£ per visit)ucNurse£22£17£34PSSRU Unit Costs 2004Assumed equivalent to district nurse/practice nurse/health visitor (including qualification costs and overheads). Uplifted for inflation.
 Cost of GP consultation (£ per consult)ucGP£23£10£30"GP surgery consult lasting 9.36mins/primary care nurse consult/GP clinic consult lasting 12.6mins
(includes qualification costs and overheads). Uplifted for inflation.
 Number of outpatient clinic visits per case treatedOP428Marchant 1998Assumptions by Marchant agreed by GDG. Upper limit is mean for 18 non-MDR cases in White and Moore-Gillon 2000.
 Visits from TB nurse per case treatedNurse636""
 GP consultations per case treatedGP001Assumption
 Cost of non-drug outpatient care (£ per active case)cOP£764£375£1,544IF(OP<1,0,IF(OP=1,ucOP1,ucOP1+(OP-1)*ucOP2))+Nurse*ucNurse+GP*ucGP
Tests
 Cost of interferon gamma test (£ per test)ucIGtest£10£5£20Internet, accessed 6/01/05Low estimate: $10 price from US (may not be appropriate for UK, and may exclude labour costs).
 Cost of culture tests (£ per test)ucCtest£7£4£11DH Tariff 2005/6Microbiology/bacteriology
 Cost of chest X-ray (£ per X-ray)ucXray£16£11£18DH Tariff 2005/6Band A. Range from 2003/4 reference costs for range.
 Interferon gamma test per case treatedIGtest001"
Culture tests per case treatedCtest426"Assumed once per clinic visit
 Chest X-ray per case treatedXray213"Assumed once every other clinic visit
 Cost of tests (£ per active case)cTest£61£18£140IGtest*ucIGtest+Ctest*ucCtest+Xray*ucX ray
MDR TB
 Proportion of active cases that are MDRpMDR1.1%0.9%1.3%HPA 2002Upper limit is for resistance to more than one first line drug
 Cost per MDR TB casecMDR£27,844£20,000£40,000White & Moore-Gillon 2000Baseline estimate is based on NHS Tariff costs (2005/6).
Total
 Cost for TB treatment for survivors (£ per case)cTBlive£5,160£1,988£11,458pMDR*cMDR+(1-pMDR)*(cTrace+cIP+cdrugs+cOP+cTests)Includes cost of contact tracing, inpatient care, outpatient care, tests and treating MDR cases
 Cost for TB treatment for fatalities (£ per case)cTBdie£4,156£1,495£9,496pMDR*cMDR+(1-pMDR)*(cTrace+cIP)Includes contact tracing, inpatient care and MDR cases
 Cost per primary case: age 10–14 (£ per case)cTB£5,157£1,928£11,598Mean treatment cost for survivors and fatalities age 10–14 (weighted by incidence and case fatality rate)
 Cost per primary case: age 15–24 (£ per case)cTB10£4,298£1,607£9,667As above, but for cases aged 15–24
 Cost per primary case: age 15–29 (£ per case)cTB15£3,891£1,455£8,751As above, but for cases aged 15–29
 Cost per secondary case (index age 10–14)cSec£5,098£1,906£11,466Mean treatment cost for secondary cases resulting from index case aged 10–14 and assuming one year time lag for transmission.
 Cost per secondary case (index age 15–24)cSec10£4,107£1,535£9,237As above but for index cases age 15–24, discounted to age 15 for index case.
 Cost per secondary case (index age 15–29)cSec15£3,719£1,390£8,364As above but for index case aged 15-29.
Prophylaxis of latent cases
Treatment for latent TB infection
 Cost of isoniazid (£ per month)ucIso£12.36£6.18£18.55See previous table
 Cost of rifampicin (£ per month)ucRif£10.76£5.45£21.51
 Cost of ethambutol (£ per month)ucEth£18.48£12.32£22.89
 Cost of pyrazinamide (£ per month)ucPyr£6.88£4.58£9.17
 Duration of isoniazid (months)IsoP666GDG recommendationRegimen for pulmonary TB (HIV-and non-MDR).
Assumes full concordance with recommended regimen. No DOTS or other increased surveillance assumed.
 Duration of rifampicin (months)RifP000"
 Duration of ethambutol (months)EthP000"
 Duration of pyrazinamide (months)PyrP000"
 Cost of treatment for latent TB infection (£ per latent case treated)cdrugsP£74£37£111(IsoP*ucIso+RifP*ucRif+PyrP*ucPyr+Eth P*ucEth)
Outpatient care
 Cost of outpatient consultation: first visit (£ per visit)ucOP1£252£214£289See previous table
 Cost of outpatient consultation: follow up visits (£ per visit)ucOP2£128£109£146"
 Cost of TB nurse home visit (£ per visit)ucNurse£22£17£34"
 Cost of GP consultation (£ per consult)ucGP£23£10£30"
 Outpatient clinic visits per case treatedOPP212Assumption by GDG
 Visits from TB nurse per case treatedNurseP006Assumption
 GP consultations per case treatedGPP001"
 Cost of outpatient care (£ per latent case treated)cOPP£379£214£668IF(OPP<1,0,IF(OPP=1,ucOP1,ucOP1+(O PP-1)*ucOP2))+NurseP*ucNurse+GPP*ucGP
Tests
 Cost of interferon gamma test (£ per test)ucIGtest£10£5£20See previous table
 Cost of culture tests (£ per test)ucCtest£7£4£11"
 Cost of chest X-ray (£ per X-ray)ucXray£16£11£18"
 Interferon gamma test per case treatedIGtestP101GDG recommendation
Culture tests per case treatedCtestP202AssumptionAssumed once per clinic visit
 Chest X-ray per case treatedXrayP102"Assumed once every other clinic visit
 Cost of tests (£ per latent case treated)ctestsP£41£0£78IGtestP*ucIGtest+CtestP*ucCtest+XrayP* ucXray
Total
 Cost per latent case detected through school BCGcPro£494£251£857cdrugsP+cOPP+cTestsPMean cost of treatment for latent TB infection for latent cases resulting from index case aged 10–14, and detected through BCG
 Cost per treated latent case (index age 15–24)cPro10£398£202£690Discounted from age 22 (mid-way for range 15–29) to age 15.
 Cost per treated latent case (index age 15–29)cPro15£360£148£504Discounted from age 22 (mid-way for range 15–29) to age 15.
Economic parameters
 Inflation 2004/5 to 2005/6Ia8.7%
 Inflation 2003/4 to 2005/6Ib14.5%DH National Tariff 2005/6 Annex B
 Discount rate (health effects)drH3.5%HMT recommended rates
 Discount rate (costs)drM3.5%"
 Exchange rate (US-UK)ER0.51Internet

Economic analysis of new entrant screening

INTRODUCTION

The UK has had a policy of screening entrants from high-risk countries for several years now through the ‘Port of Arrival’ scheme (Hogan et al 2005). New arrivals from high-incidence countries (40/100,000 or over) who are intending to stay for six months or more are identified by immigration staff and referred for initial clinical and radiographic assessment at port health control units. Local consultants in communicable disease are then notified of the results for people moving into their area and are expected to organise appropriate follow-up. In practice, follow-up is patchy, with variations in the level and type of services provided. In addition to the Port of Arrival scheme, the Home Office has more recently introduced a TB screening system for asylum seekers at fast-track induction centres. However, it is not clear whether any of these new entrant screening policies or practices represent a cost-effective use of NHS resources, since they have not been subject to formal economic evaluation.

A review of economic literature yielded only one UK-based economic evaluation of new entrant screening 351. Bothamley and colleagues appraised three screening schemes in East London: assessment at a hospital clinic for new entrants notified from the Port of Arrival scheme; screening at general practice registration; and screening of homeless people. Their results suggested that hospital-based screening of people referred from the Port of Arrival scheme appears to be cost-saving compared with no screening: the estimated cost for screening 199 people was £22,600, resulting in an estimated 9.5 cases prevented and a saving of £25,600 in potential treatment costs. However, these figures do not appear to include the costs of identification, initial assessment and notification at the port. It is also not clear how robust they are to various assumptions used to estimate the costs and savings, nor how transferable the findings are to other areas of the country with different screening systems.

There is a limited economic literature on the cost-effectiveness of TB new entrant screening overseas. For example, Dasgupta et al(2000) evaluated the cost-effectiveness of two TB screening programmes of foreign born populations and screening of resident close contacts in Montreal in 1996/7. Their results indicated that active contact tracing is cost saving, but that the costs of X-ray screening of new entrants and post-arrival surveillance are relatively high (estimated at $39,400 and $65,100 per case detected respectively). Another Canadian study (Schwartzman and Menzies 2000) used a modelling approach to estimate the cost-effectiveness of radiography and tuberculin skin testing for screening of immigrants. The authors concluded that neither method is cost-effective for low-risk groups, but that X-ray screening at or before entry may be cost-effective for high-risk groups. However, the applicability of this Canadian evidence to the UK is questionable because of obvious differences in the organisation of screening programmes and the general health care context.

Given the importance of new entrant screening in policy terms, and the lack of strong evidence on its cost-effectiveness, the GDG prioritised this as an area for economic analysis. The aims, methods and results of this analysis are presented in this Appendix.

AIMS

The cost-effectiveness of new entrant screening depends on the design and implementation of the screening programme and the context in which it is applied. It was not possible to investigate every possible permutation of the screening system within this analysis. Instead we chose to focus on four key questions:

  1. At what level of prevalence of active disease or latent infection does screening become cost-effective?
  2. What screening test is most efficient for detecting active disease in this population: a symptom checklist alone or chest X-ray combined with assessment of symptoms?
  3. What test is most efficient for identifying people with latent infection: tuberculin skin tests (TST) or interferon gamma tests (IGT)?
  4. What preventive treatments are cost-effective for this group: treatment for latent TB infection, vaccination, both or neither?

We started by building a decision tree representing an initial screening algorithm, designed as an interpretation of current screening policy. We then estimated the cost-effectiveness of some variations around this algorithm to address the above questions. Firstly, we changed the prevalence of active disease and latent infection in the population and observed the impact on the cost-effectiveness of the screening algorithm. Secondly, we investigated how the cost-effectiveness would be likely to change with different screening tests for active disease. Thus we estimated the impact of substituting X-ray screening for a simple symptom checklist for the initial identification of people with active disease at the port of arrival. Thirdly, we estimated what would happen if interferon gamma tests were to be used instead of skin tests for detection of those at risk from latent infection. In this analysis we did not include a two-stage testing strategy (TST followed by IGT) because of the greater risk of loss to follow-up in this population. Finally, we estimated how cost-effectiveness might change if we were to drop the use of prophylaxis for people with suspected latent infection and/or vaccination for young unvaccinated and uninfected individuals.

In order to estimate the cost of the screening algorithm it was necessary to make some assumptions about the organisation of services. However, we did not systematically investigate different service models. For example, we did not consider the relative costs and effects of local follow-up through hospital clinics or primary care. National and local bodies responsible for screening and follow-up services will need to consider the most efficient way to organise and deliver these services. The model also excludes other possible benefits from screening – for example, community based services could help to introduce new entrants to local health services and improve detection of other health problems. These ‘externalities’ may be an important consideration for the design of local services.

METHODS

General approach to modelling

The modelling approach taken was similar to that used for evaluation of school BCG and tests for latent infection. We used a simple decision tree to estimate the expected costs and health effects of some variations to an initial screening algorithm. The model is illustrated in Figure 1 and described in detail below. Note that, for simplicity, the dynamics of transmission within the population was not modelled. Instead, the results depend on an assumed fixed number of secondary cases per primary case. Dynamic population modelling would be expected to yield more accurate results, particularly in the longer term.

The data and assumptions used to estimate the input parameters of the model are all listed in Table 1. Parameter estimates were based on published evidence where possible. If no appropriate evidence was available, estimates were made by the guideline economist and checked by the GDG. There is a high level of uncertainty about the values for some of the model parameters. The robustness of the model results to input data and assumptions is explored in a sensitivity analysis.

Table 1. Model parameters.

Table 1

Model parameters.

The analysis is conducted according to the general principles of the NICE reference case. This includes the use of an NHS perspective, discounting of costs and effects at a rate of 3.5% per annum, incremental analysis, and sensitivity analysis. The time horizon for the model is 15 years, as this is the maximum expected period for benefit of BCG vaccination.

Population characteristics

The model estimates the costs and effects of a screening programme for a hypothetical cohort of new entrants. The cohort is characterised by an initial prevalence of active disease (pTB) and TB infection (pInfect) at entry. Those members of the cohort without active disease or latent infection have a risk of TB incidence over the model time horizon of 15 years of rNI. The risk of incident disease over this time period in those with latent infection at entry is rather higher (rLI). The proportion of the cohort unvaccinated at entry is pPrior. We also assume an age cut-off of 35 for vaccination and for treatment for latent TB infection, based on consensus view of the GDG. The proportion of the cohort aged under this cut-off is pYoung.

Screening algorithm

The initial algorithm for new entrant screening is shown as a decision tree (Figure 1). This assumes that individuals from the cohort are first invited to complete a brief questionnaire; either at the port of arrival or later at a general practice or hospital clinic. Some proportion of the total cohort (pScreen) completes the questionnaire, which has sensitivity and specificity for detecting active cases of seSQ and spSQ respectively. In addition, people aged 35 or younger are offered a tuberculin skin test (TST), which is assumed to have sensitivity and specificity for detecting TB infection of SeTST and SpTST respectively. A proportion of skin tests (pTST2) have to be repeated because they cannot be read at the appropriate time. The resulting proportion of TST results that are available, after a second attempt if necessary, is pRead.

Flowchart Icon

Figure 1. Decision tree representing baseline algorithm for new entrant screening (PDF, 22K)

Accuracy of screening and tests for latent infection

The proportion of people screening positive for symptoms is estimated as:

posScreen=(pTB*seSQ)+((1-pTB)*(1-spSQ))

The proportion of people with symptoms who actually have active TB (the ‘Positive Predictive Value’ of the symptom checklist) is:

ppvScreen=(pTB*seSQ)/(posScreen)

The proportion of people without symptoms who do not have active TB (the ‘Negative Predictive Value’) is:

npvScreen=((1-pTB)*spSQ/(1-posScreen)

The proportion of individuals without active TB who have a positive skin test is:

posTest=(pLI*seTST)+((1-pLI)*(1-spTST))

where

pLI=(pInfect-pTB)/(1-pTB)

The estimated positive and negative predictive values of the TST are:

ppvTest=(pLI*seTST)/(posTest)npvTest=((1-pLI)*spTST)/(1-posTest)

Treatment algorithm

People who screen positive for TB symptoms are invited for clinic assessment, although only some proportion of them attends (pAssess). We assume that all people with active disease who attend the clinic are accurately diagnosed (and that no people without active disease are falsely diagnosed). However, some active cases will be missed by the screening questionnaire (false negatives). These people may still be identified earlier than in the absence of screening due to a strongly positive test for TB infection. If they do not attend to have their skin test result read, or if they miss the clinic appointment, then diagnosis is delayed until clinical presentation. People without active TB are considered for preventive treatment. The algorithm assumes that prophylaxis will be offered to individuals aged 35 or younger with a positive test for latent infection. The proportion of those offered prophylaxis who start treatment is pPro. For those with latent infection, prophylaxis reduces the risk of future TB (relative risk rrPro). It is assumed that prophylaxis offers no benefits for people without current latent infection. The baseline algorithm also includes an offer of vaccination for those aged 35 or younger with a negative skin test and no evidence of prior BCG. It is assumed that a proportion pBCG of those offered vaccination accept. For uninfected individuals, vaccination reduces the risk of future disease (relative risk rrBCG). We assume that there is no benefit from vaccination of infected individuals. People with a negative diagnosis following clinic assessment for active disease are considered for treatment for latent infection or vaccination in the same way as if they had not presented with symptoms.

Health outcomes

The morbidity and mortality impact of each case of TB is estimated in terms of Quality Adjusted Life Years (QALYs). The methods used to estimate the mean QALY loss per case (qTB) are explained in the school vaccination model. For the new entrant model we used an estimate based on the expected QALY loss due to TB-related mortality and morbidity for all ages, weighted by age of TB incidence.

For each case of active TB we also assume that there are nSec secondary cases, which occur after an average delay of lagSec years. In addition to prevalent cases, the model includes QALY losses due to future incidence over the 15-year time horizon. For members of the cohort without current infection at the time of screening, we assume an average delay of lagNI years for any incident cases. For people with latent infection at the time of screening, the average time to occurrence of any incident cases is assumed to be rather lower, lagLI years. All estimated health gains are discounted to the time of screening using an annual discount rate of 3.5%, as recommended the NICE ‘reference case’ for economic evaluations. There are three mechanisms by which the screening algorithm can improve outcomes: earlier diagnosis of active disease; prevention of future disease due to prophylactic treatment of people with latent infection; and prevention of future disease by vaccination of currently uninfected people. Early diagnosis is assumed to reduce the quality of life and mortality impact of TB for the index case by qEarly QALYs. Early diagnosis is also assumed to reduce the number of secondary cases by a proportion pEarly. Cases prevented by vaccination or prophylaxis yield a QALY gain for the index case, and also for expected secondary cases.

These QALY gains due to early diagnosis and prevention will be partially offset by QALY losses due to side effects of tests or treatment. We assume fixed QALY losses for each skin test and vaccination performed (qTest and qBCG). In addition we estimate the QALY loss due to adverse reactions to prophylaxis as qPro. This figure includes an estimate of mortality due to isoniazid-related hepatitis as well as the quality of life impact of non-fatal hepatitis and other side effects. The QALY loss during treatment of active disease is already included in the qTB figure.

Costs

The model includes estimates of the cost of screening (cScreen), testing for latent infection (cTest) and reading test results (cRead), clinic assessment for suspected cases (cAssess), vaccination (cBCG), prophylaxis (cPro) and management of active disease (cTB). These cost estimates are the same as in the other two economic models. Note that for people with active disease, the model does not include an additional cost for clinic attendance, since the cost cTB already includes initial costs of assessment. Where appropriate, costs are discounted using an annual rate of 3.5%.

RESULTS

Cost-effectiveness of baseline algorithm

Under baseline assumptions the screening algorithm does not appear to be cost-effective (see Table 2). With 0.5% active and 20% latent TB in the cohort at the time of screening, the algorithm is estimated to cost over £250,000 per QALY gained, which far exceeds the usual NICE threshold of £20–30,000. This result was robust to changes in the input parameters. Each parameter was varied, one at a time, from the lower to the upper limit shown in Table 1 (a simple ‘one-way’ sensitivity analysis). For all except three parameters (pTB, nSec and rLI) the estimated cost per QALY of the baseline algorithm remained above £200,000, and in no case did it fall below £30,000 per QALY.

Table 2. Cost-effectiveness of baseline algorithm.

Table 2

Cost-effectiveness of baseline algorithm.

The cost per QALY fell to about £39,000 when the prevalence of active TB in the cohort at the time of screening (pTB) was increased to our upper limit of 5%. For the algorithm to meet a £30,000 cost-effectiveness threshold, the prevalence of active TB in the cohort would have to be greater than 6% (see Table 3). To meet a more stringent cost-effectiveness threshold of £20,000 per QALY, the prevalence would have to be over 8%.

Table 3. Sensitivity of algorithm cost-effectiveness to TB prevalence.

Table 3

Sensitivity of algorithm cost-effectiveness to TB prevalence.

The next most sensitive input parameter was the number of secondary cases per primary case (nSec). At our upper limit of 0.75 the estimated cost per QALY for the baseline algorithm was about £91,000. There would have to be over 2 secondary cases per primary case to reach a cost-effectiveness threshold of £30,000 per QALY. The 15-year incidence of TB in people with latent infection at the time of screening (rLI) would have to rise from a baseline estimate of 2% to about 18% to meet a cost-effectiveness threshold of £30,000 per QALY.

Cost-effectiveness of radiographic screening for active disease

In the baseline model the symptom checklist used to identify people with suspected TB for further investigation is assumed to have a sensitivity of 80% and specificity of 70%. Given the expected prevalence of 0.5%, this means that only about 1% of people with symptoms would actually have the disease, and that over 99% of those without symptoms would not have the disease. Radiographic screening is more expensive than symptom screening alone (about £16 more per person), but if is also expected to be more accurate. In our baseline model we assumed that chest X-rays would have a sensitivity of 95% and specificity of 98% in this population. If true, this would improve the targeting of further investigation: about 20% of those with a suspicious X-ray would have TB and over 99% with a clear X-ray would not have TB. This would reduce the overall cost of the screening programme by about £10 per patient and give a slight improvement in health outcomes.

The superiority of radiographic screening is robust to changes in its sensitivity. Even if the sensitivity of chest X-ray for active TB were only 50% it would remain highly cost-effective (the symptom checklist would cost over £1m for each additional QALY gained). This is not surprising given the low baseline estimate of prevalence in this population, and hence the limited scope for gain by reducing false negatives. The results depend more on the avoidance of false positive results, and hence on the relative specificities of the screening methods. Provided that the specificity of X-ray screening is no lower than 78.5% it remains more cost-effective than purely symptomatic screening. The result is also quite robust to changes in the relative cost of radiographic and symptomatic screening: as long as the additional cost of the chest x-ray is no more than £50 greater then it remains cheaper overall in the base case.

Cost-effectiveness of Interferon Gamma test for latent infection

The baseline results reported above depend on an assumed sensitivity of 90% and specificity of 80% for the skin test. With 20% prevalence of infection this implies that only 53% of those with a positive skin test will be infected and that 97% of those with a negative skin test are not infected. Although it costs an estimated £15 more per person, the interferon gamma test may still be more cost-effective than skin testing in this context if it offers a sufficient improvement in specificity and hence better targeting of preventive treatment. In our baseline model, where we assume that the interferon gamma test has a sensitivity of 90% and specificity of 80% (10% lower than the skin test), it appears to be cost-saving: saving about £4 per person and slightly improving expected health outcomes. In fact, the interferon gamma test appeared to be cost-effective provided that its specificity was no less than 4% better than the skin test. At 90% specificity, the interferon gamma test is cost-effective provided that it costs no more than about £45. At 85% specificity a cost of only £27 could be justified.

Cost-effectiveness of vaccination

The model predicts that vaccination is a cost-effective component of a new entrant screening programme. Removing vaccination from the baseline algorithm led to a small increase in overall NHS costs and a small reduction in health gains: an extra cost of about £200 and loss of about 0.02 of a QALY per 1,000 offered screening. The result was robust under one-way sensitivity analysis. The cost of the BCG would have to be over £27 per patient before vaccination drops below the cost-effectiveness threshold of £30,000 per QALY. Vaccination remained cost-saving in all other scenarios tested.

Cost-effectiveness of prophylaxis

The use of prophylaxis for people with suspected latent TB infection was not supported by this model. Under the baseline analysis it was estimated to cost an extra £400,000 per QALY gained. It remained highly cost ineffective under sensitivity analysis. The parameter with the biggest impact of the cost-effectiveness of prophylaxis was the future risk of TB in people with latent infection at screening (rLI). This had to rise to over 12% over the fifteen year time horizon of the model before prophylaxis appeared to be cost-effective.

Comparison of screening strategies

Finally, we compared the cost-effectiveness of eleven possible strategies, based on permutations of:

  1. The method for screening for active disease - none, symptoms (SQ), or x-ray (XR);
  2. The method of screening for latent infection - none, TST or IGT;
  3. The preventive interventions offered - none, BCG and/or prophylaxis.

The results of this analysis under the baseline parameter values are shown in Table 4. Applying a standard £20–30,000 cost-effectiveness threshold, this suggests that the optimum strategy is ‘no screening’.

Table 4. Baseline comparison of 11 screening strategies.

Table 4

Baseline comparison of 11 screening strategies.

This result is robust to changes in all parameters except two (nSec and pTB). Firstly, if transmission of TB is higher than expected (greater than about 0.4 secondary cases per primary case) then X-ray screening and skin testing followed by BCG if appropriate becomes cost-effective. At even higher levels of transmission (above about 1.2 secondary cases per primary case) substitution of interferon gamma testing for skin testing appears cost-effective.

Secondly, the optimum strategy is sensitive to the baseline prevalence of TB in the cohort at the time of screening. This is illustrated in Table 5. Strategies that are subject to either simple or extended dominance have been removed from this table. It can be seen that at low levels of prevalence none of the screening strategies is cost-effective. For populations with a high prevalence of active TB (above about 100 in 10,000) radiographic screening, and possibly skin testing followed by vaccination if appropriate, starts to become cost-effective.

Table 5. Cost effectiveness of non-dominated strategies by prevalence.

Table 5

Cost effectiveness of non-dominated strategies by prevalence.

CONCLUSIONS

A decision analytic model was used to estimate the cost-effectiveness of alternative screening algorithms for new entrants from high-risk countries. The economic model was based on an initial algorithm which included initial screening for active disease using a symptom checklist with clinic follow-up for suspected cases, skin testing for detecting latent infection in new entrants aged 35 or younger. It was assumed that prophylaxis would be offered to those with positive skin tests, and no active disease, and that BCG vaccination would be offered to people with an negative skin test and no evidence of prior BCG. The model included assumptions about the attendance and treatment concordance rates. We then estimated the cost-effectiveness of variations to the screening algorithm, and the overall cost-effectiveness of the algorithm as a function of the prevalence of active and latent TB in the cohort, and the future incidence for people with/without latent infection at the time of screening.

The model used a simple decision tree approach, assuming a fixed number of secondary cases per primary case, rather than modelling the dynamics of transmission within the population. The results should thus be treated with caution. Caution is also required because of considerable uncertainty over various data inputs and assumptions, and also because of likely variation in programme effectiveness and costs in different areas. As far as possible, the model was based on best available empirical evidence. However, no data were available for some key parameters, so judgement from GDG members was used to estimate likely ranges of values.

It is important to recognise that the model does not take account of other potential benefits of screening – for example, community based screening may act to introduce new entrants to local health services, and as a screen for other possible health problems. The model also does not take account of other ways in which screening and treatments could be better targeted. For example, the decision to offer prophylaxis could be informed by the individuals’ likely exposure to TB, their social environment, and/or indicators of latent infection from Xray. The economic model suggests that prophylaxis is not cost-effective in the context of new entrant screening. Using the basecase assumptions, the estimated incremental cost per quality adjusted life year gained for including prophylaxis in the new entrant screening algorithm was nearly £400,000. This result was robust to variation in the model parameters.

The model predicts that BCG vaccination is cost-saving for the NHS in the context of new entrant screening. Removing vaccination for TST negative new entrants from the new entrant screening algorithm led to a cost saving of £20,000 and a QALY gain of 1.8 per 100,000 screened, under the basecase assumptions.

The cost-effectiveness of initial screening for active disease with a symptom checklist compared with chest X-ray depends on their relative costs and accuracies. Under the basecase assumptions, the model suggests that although X-ray screening is more expensive, it leads to an overall saving in NHS expenditure due the lower number of false positive results that is predicted.

The model suggests that, despite its higher initial cost, interferon gamma testing might be a cost-effective alternative to skin testing if it is demonstrated to give a lower number of false positive results. Under the basecase assumptions, the model predicted that interferon-gamma tests would be cost-saving in comparison with skin tests. At low levels of prevalent TB in the cohort tested, none of the screening algorithms was cost-effective. The algorithm without prophylaxis achieves an incremental cost-effectiveness ratio (ICER) of £30,000 per QALY at a TB prevalence of about 3%, and an ICER of £20,000 per QALY at about 4% prevalence. This is relatively high compared with rates of disease found in many new entrant screening programmes. If a more accurate method of screening for active disease (such as chest X-ray) is substituted for simple symptomatic screening, screening becomes cost-effective at lower levels of prevalence (at about 1% or higher).

Copyright © 2006, Royal College of Physicians of London.

All rights reserved. No part of this publication may be reproduced in any form (including photocopying or storing it in any medium by electronic means and whether or not transiently or incidentally to some other use of this publication) without the written permission of the copyright owner. Applications for the copyright owner’s written permission to reproduce any part of this publication should be addressed to the publisher.

Bookshelf ID: NBK45809

Views

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...