Validation of Artificial Intelligence Enabled TB Screening and Diagnosis in Zambia

Tuberculosis (TB) is a global epidemic and for many years has remained a major cause of death throughout the developing world. Zambia is among the top 30 TB/HIV high burden countries. Chest X-ray (CXR) is recommended as a triaging test for TB, and a diagnostic aid when available. However, many high-burden settings lack access to experienced radiologists capable of interpreting these images, resulting in mixed sensitivity, poor specificity, and large inter-observer variation. In recognition of this challenge, the World Health Organization has recommended the use of automated systems that utilize artificial intelligence (AI) to read CXRs for screening and triaging for TB. In this study, we primarily evaluate the performance of our AI algorithm for TB, and secondarily for Abnormal/Normal.

Study Overview

• Status: Completed
• Conditions: Tuberculosis

• Study Type: Observational
• Enrollment (Actual): 2432

Contacts and Locations

Study Locations

• Zambia
  • Lusaka, Zambia, 10101
    • Chainda South Health Facility
  • Lusaka, Zambia, 10101
    • Chawama first level hospital
  • Lusaka, Zambia, 10101
    • Kanyama level 1

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 18 years and older (Adult, Older Adult)
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

The study will be conducted from Chainda South Clinic, Chawama and Kanyama General Hospitals. These facilities are selected based on existing access to digital radiography. Participants will be drawn from patients attending these health facilities for:

• Health services
• TB households of those who are close contacts of TB households within the catchment area of these facilities

Description

Inclusion Criteria:

• Participants who are 18 years and older with a known HIV status or are willing to undergo HIV testing if unknown HIV status and meet the following criteria will be included in the study:

  • Presumptive TB patients defined as having any of the following:

    ○ Cough, Weight loss, Night sweats, Fever

  • Household /close TB contacts regardless of symptoms
  • Newly diagnosed HIV regardless of symptoms.

Exclusion Criteria:

• Individuals who do meet the above inclusion criteria will be excluded. In addition, individuals with history of TB treatment within 365 days prior to enrolment will be excluded.

Study Plan

How is the study designed?

Number of groups / cohorts

2

Cohorts and Interventions

Group / Cohort
Pilot Group to calibrate the operating points for AI algorithms (Estimated Enrollment up to 500); Diagnostic Test: TB AI algorithm performance in detecting active TB. Diagnostic Test: TB diagnosis from sputum and urine (Smear microscopy, Xpert MTB RIF/ultra, Lipoarabinomannan (LAM) and mycobacterial culture) Diagnostic Test: Abnormal/Normal AI algorithm to detect abnormal/normal CXRs. Diagnostic Test: Radiologist evaluation of CXRs for active TB, abnormal/normal. Diagnostic Test: Labs: Hemoglobin level, HIV status, CD4 count.
Main Cross Sectional Group (Estimated Enrollment 1932 minus the volume in pilot); Diagnostic Test: TB AI algorithm performance in detecting active TB. Diagnostic Test: TB diagnosis from sputum and urine (Smear microscopy, Xpert MTB RIF/ultra, Lipoarabinomannan (LAM) and mycobacterial culture) Diagnostic Test: Abnormal/Normal AI algorithm to detect abnormal/normal CXRs. Diagnostic Test: Radiologist evaluation of CXRs for active TB, abnormal/normal. Diagnostic Test: Labs: Hemoglobin level, HIV status, CD4 count.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Pilot Group to calibrate the operating points for AI algorithms	1. Operating point selection for TB AI algorithm and Abnormal/Normal AI algorithm on CXRs for outcomes listed in Main Cross Sectional Group.	2 months
Main Cross Sectional Group	1. TB AI algorithm sensitivity and specificity in detecting active TB on CXR compared to panel of radiologists	7 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Main Cross Sectional Group:	1. TB AI algorithm sensitivity and specificity in detecting active TB compared to World Health Organisation (WHO) performance guidelines of 90% sensitivity and 70% specificity	7 months
Main Cross Sectional Group	2. Abnormal/Normal AI algorithm sensitivity and specificity compared to 90% sensitivity and 50% specificity.	7 months

Collaborators and Investigators

• Sponsor: Centre for Infectious Disease Research in Zambia

Study record dates

Study Major Dates

• Study Start (Actual): November 22, 2021
• Primary Completion (Actual): September 30, 2022
• Study Completion (Actual): November 30, 2022

Study Registration Dates

• First Submitted: November 18, 2021
• First Submitted That Met QC Criteria: November 18, 2021
• First Posted (Actual): December 1, 2021

Study Record Updates

• Last Update Posted (Actual): May 15, 2025
• Last Update Submitted That Met QC Criteria: May 13, 2025
• Last Verified: May 1, 2025

More Information

Terms related to this study

• Keywords: Screening; Diagnosis; Artificial Intelligence; Chest X-ray
• Additional Relevant MeSH Terms: Infections; Gram-Positive Bacterial Infections; Bacterial Infections; Bacterial Infections and Mycoses; Actinomycetales Infections; Mycobacterium Infections; Tuberculosis
• Other Study ID Numbers: Google AI

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No