Screening of immigrants in the UK for imported latent tuberculosis: a multicentre cohort study and cost-effectiveness analysis

Summary Background Continuing rises in tuberculosis notifications in the UK are attributable to cases in foreign-born immigrants. National guidance for immigrant screening is hampered by a lack of data about the prevalence of, and risk factors for, latent tuberculosis infection in immigrants. We aimed to determine the prevalence of latent infection in immigrants to the UK to define which groups should be screened and to quantify cost-effectiveness. Methods In our multicentre cohort study and cost-effectiveness analysis we analysed demographic and test results from three centres in the UK (from 2008 to 2010) that used interferon-γ release-assay (IGRA) to screen immigrants aged 35 years or younger for latent tuberculosis infection. We assessed factors associated with latent infection by use of logistic regression and calculated the yields and cost-effectiveness of screening at different levels of tuberculosis incidence in immigrants' countries of origin with a decision analysis model. Findings Results for IGRA-based screening were positive in 245 of 1229 immigrants (20%), negative in 982 (80%), and indeterminate in two (0·2%). Positive results were independently associated with increases in tuberculosis incidence in immigrants' countries of origin (p=0·0006), male sex (p=0·046), and age (p<0·0001). National policy thus far would fail to detect 71% of individuals with latent infection. The two most cost-effective strategies were to screen individuals from countries with a tuberculosis incidence of more than 250 cases per 100 000 (incremental cost-effectiveness ratio [ICER] was £17 956 [£1=US$1·60] per prevented case of tuberculosis) and at more than 150 cases per 100 000 (including immigrants from the Indian subcontinent), which identified 92% of infected immigrants and prevented an additional 29 cases at an ICER of £20 819 per additional case averted. Interpretation Screening for latent infection can be implemented cost-effectively at a level of incidence that identifies most immigrants with latent tuberculosis, thereby preventing substantial numbers of future cases of active tuberculosis. Funding Medical Research Council and Wellcome Trust.

from India (TB incidence 170/100000) but screening is limited to individuals arriving from countries with a TB incidence ≥200/100,000. 2. Immigrants with LTBI who screen positive but decline to commence chemoprophylaxis. 3. Immigrants with LTBI who screen positive, accept chemoprophylaxis, do not develop hepatotoxicity, complete therapy but it is not effective. 4. Immigrants with LTBI who screen positive, accept chemoprophylaxis, do not develop hepatotoxicity do not complete therapy which is not effective. 5. Immigrants with LTBI who screen positive, accept chemoprophylaxis, develop hepatotoxicity which resolves, complete therapy but it is not effective. 6. Immigrants with LTBI who screen positive, accept chemoprophylaxis, develop hepatotoxicity which resolves, do not complete therapy but it is not effective. 7. Individuals with LTBI who screen positive, accept chemoprophylaxis, develop hepatotoxicity which does not resolve resulting in them stopping therapy early. This renders them still latently infected. 8. Immigrants with LTBI but who actually test false-negative with the IGRA.
These individuals remain at risk of progressing from the latent state to active TB disease at a fixed rate. In the absence of reliable data about the proportion of migrants with HIV infection it is assumed that none of the immigrants have HIV.
If an individual with LTBI breaks down to active TB disease all strains are assumed to be fully drug sensitive. Individuals with active TB are modelled to have a fixed number of contacts which will result in a fixed number of secondary active TB cases and LTBI. Depending on the severity of disease, a proportion of individuals will need to be hospitalised whilst the remainder will be managed as outpatients. It is assumed that all subjects accept treatment and that treatment for all cases of active TB follows national guidelines with compliance, and cure, fixed at 100%. Once an individual has been treated for active TB they cannot be re-infected during the course of the 20 year model. In view of the low mortality rate from TB in the UK it is assumed that there is no TB/background mortality during the 20-year horizon of the model.

Input parameters and probabilities
Input data were obtained from the present multi-centre study whilst probabilities for transitioning between states were obtained from previous literature (see online supplementary table 2).
Key to the cost-effectiveness analysis were the parameters used to describe the performance of the IGRA (QuantiFERON-Gold In-tube) in diagnosing LTBI. These were obtained from the most recent meta-analysis on IGRA performance which concluded the QuantiFERON Gold In-tube has a specificity of 99% and sensitivity, in developed countries, of 84%. 5 The high specificity means that the proportion of false-positive results is relatively small, given the relatively high prevalence of LTBI in the cohort. 4 6 In contrast, the sensitivity of the IGRA impacts on the proportion of false-negative results.
When calculating the true prevalence of LTBI in the tested cohort, it is important to take into account test performance. If, for example, 20% of individuals are IGRA positive it would be incorrect/inaccurate to simply assume that these 20% represent all truly infected individuals. The reason for this is that, depending on test sensitivity and specificity, a proportion of positives will be falsely-positive whilst some negatives will be falsely-negative. We therefore calculated the true prevalence of LTBI in the cohort by using the following formula:

Prevalence of LTBI = Probability of a positive result -(1-Test specificity)/((Test sensitivity)-(1-Test specificity))
This is important because, returning to our example of 20% of individuals being IGRA positive when test performance suggests a sensitivity of 84% and specificity of 99%, the true prevalence is actually 22.9%.
Another important, but poorly understood, parameter was the rate at which immigrants with LTBI reactivated and progressed to active TB disease. Although immigrants should, in theory, have a lower rate of progression than recent contacts of smear-positive tuberculosis, it could be argued that those individuals arriving from high TB burden countries are, in fact, akin to recent contacts as they will have been recently and repeatedly exposed to individuals with infectious tuberculosis. This makes it difficult to parameterise the progression rate with full certainty. For example whilst Marks et al calculated 6.7% progression over a 40 year period in TST positive (>15mm) Southeast Asian refugees 7 , data from the UK, in a predominantly Southeast Asian population suggests that over a 10 year period approximately 13% of TST positive, untreated, immigrants (primarily from the Indian Subcontinent) will go on to develop active TB. 8 Horsburgh estimated that in [16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35] year olds with a >15mm TST (not recently converted) the annual risk of reactivation was 0.19%. 9 If the skin test was >15mm and there was recent conversion then the annual risk of reactivation would 0.56%. 9 In view of the large difference in published data we assumed that 5% of the cohort with LTBI, in the absence of chemoprophylaxis, would progress to active TB over the 20 year time horizon; a suitably wide range was explored in the sensitivity analysis.

Costs
Component costs considered were primarily direct costs obtained from economic evaluations conducted for the UK NICE TB guidelines 1 , and its forthcoming update, uplifted to 2010 prices(see table 2 for costs) using the Consumer Prices Index. In the present analysis, indirect costs such as transportation and loss of earnings by patients were not considered. Both costs and non-monetary health effects were discounted at an annual rate of 3.5%, which reflects UK Treasury and NICE recommendations. 10 11

Effects
The main effects considered in the model were the number of cases of active tuberculosis that would be predicted to occur over the 20-year time horizon and the number needed to treat (in other words the number of individuals that need to be treated for LTBI) to prevent one case of active TB.

Cost-effectiveness
As recommended by the Panel of Cost-effectiveness in Health and Medicine, the comparative performance of the different screening protocols was measured using the Incremental Costeffectiveness ratio (ICER -see equation 1) which quantifies the trade-offs between switching from one competing, mutually-exclusive, intervention to another. 12 The higher the ICER, the less cost-effective the intervention is.

Sensitivity analysis
Parameter uncertainty can potentially affect the results of the cost-effectiveness analysis. A simple oneway sensitivity analysis was therefore undertaken to explore the impact that changes in all key parameters and costs had on the number of cases of active TB occurring over 20 years, the costs and the associated ICERs.    figure 1a. Decision tree used for the health economic analysis (Individuals fully cured of LTBI are assumed to remain free of further infection for the 20-year time horizon of the model. *For clarity all "reactivate to active TB subtrees" are shown in figure 1b below). Please note that as data is available on all migrants we are able to compute, at each incidence threshold, the number of migrants screened/not screened and subsequently the proportions that are IGRA positive and IGRA negative).

Supplementary table 1. Screening thresholds considered in the cost-effectiveness analysis
Supplementary figure 1b. Decision subtree used to describe the events that occur if an individual reactivates to active TB (although we have shown a branch/node for not accepting treatment for active TB, in the model we assume that all individuals do accept treatment for active TB) Footnotes: Individuals who do not accept treatment(*) are represented in the model for completeness but, in reality, it is assumed that no individuals refuse treatment for active TB (ie. this is a terminal node) Individuals who do not complete treatment(**) are represented in the model for completeness but, in reality, it is assumed that all patients complete treatment Individuals who are active TB not cured(***) are represented in the model for completeness but, in reality, it is assumed that as they have all completed treatment none of them move into the active TB not cured group