is a personalised risk predictor for incident tuberculosis (TB) for use in settings with low TB incidence.
Your browser does not support SVG

After entering the required inputs, the tool uses a validated model to calculate personalised estimates of an individual’s risk of developing TB over the next 2 years (from the point of latent TB testing), according to whether or not they receive preventative treatment. The estimates are designed to facilitate shared decision making between patients and clinicians when considering initiation of TB preventative treatment.
Your browser does not support SVG

PERISKOPE-TB is designed for use by clinicians when assessing people tested for latent TB infection, using either a QuantiFERON test, T-SPOT.TB or the Mantoux tuberculin skin test. The model is only validated for use in settings with annual TB incidence ≤20/100,000 persons. Further information on country-level TB incidence can be found below. The model assumes that the person being assessed does not have evidence of TB disease, and that the latent TB test has been performed ≥6 weeks after the individual’s last exposure to TB (including contact with an index case or migration from a country with high TB incidence). We define ‘preventative treatment’ as any regimen approved by WHO guidance.
Your browser does not support SVG

The research underlying development and validation of PERISKOPE-TB is published in Nature Medicine, and is accessible here.

PERISKOPE-TB was developed through a large international collaboration, led by researchers at University College London. We pooled individual level data from 15 previous studies done in 20 countries. These studies tested people for latent TB infection, and then followed them up to identify which participants developed TB disease. These datasets were used to develop the prediction model, and to validate its use for clinical practice. Full details of the PERISKOPE-TB model - including development methodology, the final model parameters and a list of contributors - are available in our peer-reviewed publication.

The user is required to enter age, latent TB test result, history of TB contact, country of birth, HIV status and history of transplant receipt in order to generate a personalised TB risk prediction. At least one valid (non-indeterminate) latent TB test result (either QuantiFERON test, T-SPOT.TB or the Mantoux tuberculin skin test) is required. Quantitative test results are preferred, though a binary (positive/negative) result can be entered for the QuantiFERON or T-SPOT.TB if no quantitative value is available. In the latter case, an ‘average’ positive or negative value will be entered into the model, as appropriate. If >1 latent TB test has been done, the QuantiFERON result will enter the model in preference to T-SPOT.TB, while the tuberculin skin test will only be used if no valid interferon gamma release assay result is available.

For recent TB contacts, the user must enter whether the index case is a household or non-household contact. If household, the index case sputum smear status is required. If this is not known, we advise that users input positive and negative sputum smear statuses to generate separate predictions, in order to assess the range of predicted risk.

Assumptions and supporting data are described in detail in our peer-reviewed publication and include:

  • - QuantiFERON (Gold-in-Tube and Gold-Plus), T-SPOT.TB and the tuberculin skin test are assumed to have equivalent prognostic value when transformed to a percentile scale.
  • - Preventative treatment regimens approved by WHO guidance are assumed to have equal effectiveness, including among all patient subgroups.
  • - People are defined as migrants from a country with high TB incidence if the annual country incidence is ≥100/100,000 in the year of migration.
  • - A single latent TB test result enters the model to generate the predictions. For people who have undergone serial testing, clinical judgement should be used when interpreting the risk estimates to account for longitudinal changes.
  • - For people tested due to recent contact, it is assumed that screening is done soon after exposure. For contacts who are screened later (e.g. >1 year from exposure), clinical judgement should be applied when interpreting the estimates since risk generally declines with increasing time since exposure.

The PERISKOPE-TB model is developed and validated for use in settings with annual TB incidence ≤20/100,000 persons. The world map below shows countries that meet these criteria in green, based on 2018 data. Country-level annual TB incidence estimates are available here.

The PERISKOPE-TB model is a flexible parametric survival model that includes the following predictor variables: age, normalised percentile latent TB test result, a composite ‘TB exposure’ variable (household contact of smear positive index case; other contact; migrant from high TB burden country with no contact; or no exposure), time since migration for migrants from high TB burden settings, HIV status, receipt of a solid organ or haematological transplant, and commencement of preventative treatment. Age and normalised latent TB test result are modelled as continuous variables, using restricted cubic splines, while the ‘TB exposure’ variable is modelled with time-varying covariates. Therefore, the best way to demonstrate how the predictor variables are weighted in the final model is to show this visually. Plots showing associations between each predictor and incident TB risk are available here.

A full description of the methods underlying model development and the final model parameters can be found in our peer-reviewed publication.

PERISKOPE-TB was developed and validated using individual level data from >30,000 adults and children in 20 countries, thus ensuring that the model is directly data-driven and widely generalisable to adult and paediatric populations in low TB transmission settings worldwide. Other models developed and validated in Peru have sought to estimate personalised (Saunders et al., 2017; Li et al., 2020) or household-level (Saunders et al., 2019) TB risk among recent case contacts. One existing model (TSTin3D) aims to estimate TB risk among people tested for latent TB in all settings, but has not been publicly validated, and is not based on individual level data. Instead, it is parameterised mathematically from multiple sources using aggregate data.

We based PERISKOPE-TB entirely on the underlying datasets to ensure that it is fully data-driven. Thus, during model development, we only included variables that were available for most participants in the underlying datasets. Therefore, data on recent TST/IGRA conversion (identified through serial testing), diabetes, nutritional status, chest radiographic fibrotic lesions, smoking, and other immunosuppressive conditions (such as renal disease, corticosteroid use or biologic agents) that may be associated with increased risk of TB are not included in the model. However, the model presented is likely to include most of the strongest co-variate predictors and can be iteratively updated as new datasets emerge. The model also does not require gender or BCG vaccination status to be entered, since these variables were not found to influence TB risk in our analysis. The model is only validated for use in settings with annual TB incidence below 20/100,000 persons.

We plan to continually update the PERISKOPE-TB model to refine the parameters as more datasets become available. In addition, we hope to integrate personalised risks of drug toxicity during preventative treatment to further inform risk/benefit decisions regarding treatment initiation. Finally, we also plan to generate a patient-facing interface that provides information directly to patients, in order to further facilitate shared decision making for preventative treatment initiation. Prior to launching this, we will consult widely with patient and public groups. Please contact us if you have further suggestions on how we may improve the tool.

Our website was built by researchers from University College London using OpenStack, Shiny Server and Bootstrap. Our logo and landing page were designed by Paul Donnelly.

Clinical judgement must be applied when using PERISKOPE-TB, and clinicians remain responsible for all clinical decisions and their consequences. Every effort has been made to ensure that the model underlying PERISKOPE-TB is robust and the methods have been subject to rigorous academic peer review.

The content of PERISKOPE-TB is protected by copyright law. The tool is freely available for clinical use, but must not be used for commercial purposes under any circumstances. By using this tool, it is assumed that you agree to these conditions.

Please email us for any general enquiries or technical issues.

The research underpinning PERISKOPE-TB was funded by the National Institute for Health Research (NIHR).

The content of PERISKOPE-TB is protected by copyright law.