FeedbackThe choice of screening and diagnostic algorithms should be based on:
- the specific objectives of screening;
- the accuracy and yield of the screening and diagnostic tests (see table of modelled performance in Annex 2);
- the profile of the prioritized risk groups;
- the TB prevalence in the risk groups;
- the costs, availability and feasibility of different tests; and
- the ability to engage the population to be screened.
The specific objectives of screening partly determine the relative importance of the sensitivity of the algorithm as compared with its specificity, as well as the trade-off between cost and yield or potential epidemiological impact. For example, if one objective is to determine eligibility for TPT (for example, as part of an investigation of contacts, people living with HIV or other populations or individuals who may benefit from TPT), it is critical to have very high sensitivity (and thus very high negative predictive value of a test result), even if the specificity is suboptimal (which in this case might lead to referral of additional people for diagnostic evaluation and possibly unnecessary treatment for TB disease). In other situations, it may be critical to avoid false-positive diagnoses and maximize efficient use of limited resources for diagnostic evaluation, and a less sensitive but highly specific algorithm might be preferable, such as a clinic-based screening programme in a densely populated urban area, in which laboratory capacity and the supply of cartridges for diagnostic testing would be rapidly depleted if a screening and diagnostic algorithm with low specificity was used.
The profile of the risk group can influence the choice of algorithm because the accuracy of certain tools is affected by underlying biological factors associated with certain risk factors (for example, CXR screening is less sensitive in people living with HIV). Certain considerations for the best algorithms for specific risk groups are based on their risk for TB and for unfavourable outcomes if TB is not detected early and logistical considerations in screening specific to the risk group and the location in which screening is conducted (see further discussion below).
The prevalence of TB in a risk group directly affects the predictive values of all tests and therefore the occurrence of true or false results. The lower the prevalence, the more important it is that the algorithm has very high specificity, in order to avoid a high proportion of false-positive diagnoses.
The total cost of an algorithm depends on the unit cost of each test (including start-up and running costs), the total number of tests required and the overhead costs for delivering the services. Different algorithms require different numbers of tests for any given population with any given TB prevalence. The tables in Annex 2 provide the estimated numbers of tests required for different algorithms in relation to case-detection yield. The tool described in 3.3 can be used to generate cost estimates for each algorithm and risk group according to local cost assumptions. This information can be used to conduct a simple cost–effectiveness analysis of the cost per true case detected. The availability, cost and feasibility of tests may, however, differ considerably in different parts of the health-care system. Outreach screening requires consideration of mobility and field conditions. For example, digital CXR technology offers lower running costs and greater mobility than conventional CXR but requires a high initial investment. Symptom screening may be relatively low cost, especially in integrated services, but it is also relatively insensitive. Diagnostic evaluation may become more feasible under outreach conditions if proper sputum collection and transport can be organized. The additional resources required to implement TB screening should not discourage managers from investing in this intervention but should stimulate mobilization of the necessary funds.
Consideration must be given to the ability to engage with the population to be screened. Although the algorithm used will have significant implications for the budget and logistics, so too will the approach used to conduct screening. Contact investigation might require home visits, or individuals with TB can be requested to bring their contacts to a health facility to be tested. Although the latter option may be far cheaper, far fewer people may actually be screened. Similarly, community outreach may involve setting up mobile treatment teams and laboratories, home visits or simply using loudspeakers to announce the availability of testing services. Different approaches work differently in different settings, and their impact will depend on the number of people reached and tested and on the yield. The acceptability of a given test and the beliefs of people who are screened and healthcare workers may have to be considered. Out-of-pocket expenditure required to complete screening should also be considered.
Considerations for algorithms for risk groups
The prevalence of TB and risks of poor health outcomes or mortality, logistical factors associated with the likely location of screening and considerations for initiating TPT for certain risk groups all influence the choice of screening algorithm. Certain algorithms inevitably require more resources, and therefore resource availability will likely determine which algorithm is feasible.
Contacts
As close contacts of individuals with TB have a high prevalence of TB, their high risk of TB and their eligibility for TPT indicate urgent screening of this risk group. As the goal of screening in this group is to identify TB disease early and to rule out TB accurately, a highly sensitive algorithm is preferred – if possible one that begins with CXR because of its high sensitivity and specificity. Screening of contacts should ideally begin in the patient’s household to ensure high coverage of this risk group. Thus, either transport of the patient to a nearby health facility or mobile CXR will be required to implement CXRbased algorithms in this risk group. The cost of such screening will be substantial, but this risk group is smaller than other groups. Although a CXR-based algorithm is preferred for this group, a more feasible algorithm must be selected when CXR services are not yet available for the screening programme.
Miners
A CXR-based screening approach, together with screening for symptoms of TB and lung disease, is also preferred for miners exposed to silica, given their high risk of lung disease (including TB) and lung damage from silicosis. Large mines often have facilities on site to conduct CXR screening for employees; smaller, informal mines may have limited capacity and may have to use other providers while increasing capacity.
Prisoners
Given the high risk of transmission in this group, a highly sensitive algorithm beginning with CXR is preferred. Larger prisons and penitentiary institutions may have radiography capacity on site or can bring mobile vans for screening campaigns. In smaller institutions or locations where CXR capacity is not available, screening algorithms based on symptoms or mWRD may be acceptable until CXR services are available.
People with clinical risk factors
In settings where the general TB prevalence is > 100/100 000, TB screening may be conducted among people with TB risk factors who are seeking health care for any medical reason or among those who are in health care. Access to radiography is more likely in a health facility. This can maximize screening sensitivity. Symptom screening is also valuable for immediate decisions on triage and infection control.
General population and communities with structural risk factors
For screening in the community, in populations with structural risk factors for TB and/or in the general population when the TB prevalence is ≥ 0.5%, a highly sensitive screening algorithm will provide the highest yield in terms of maximizing case detection, as substantial work is usually required to take intervention activities into the field. Such an algorithm, however, requires substantial resources for implementation. Screening for symptoms is much easier but is less sensitive and specific, depending on the symptom approach, and has a smaller potential impact on population prevalence or transmission. Screening with mWRDs is highly accurate (particularly specific) but has substantial resource implications.