Sensitivity and specificity of tuberculosis screening tools in people with diabetes

In a study of adults with diabetes in South Africa, symptom-based screening for tuberculosis demonstrated poor sensitivity, while chest X-ray screening offered improved but still suboptimal sensitivity and reduced specificity in individuals with a history of TB.

The Gist

Imagine you are running a security checkpoint at a busy train station. Your job is to find a very specific, dangerous intruder (let's call him "TB") hiding among thousands of regular commuters. However, this station has a special group of travelers: people with "Diabetes." The rules say you must check these specific travelers extra carefully because they are more likely to be hiding the intruder.

The big question this study asked was: Are our current security tools actually good at finding the intruder in this specific group of people?

Here is what the researchers found, explained simply:

The Setup: The "Diabetes" Train Station

The researchers set up a study in South Africa, a place where the "intruder" (TB) is very common. They looked at 673 adults with diabetes who were already visiting their doctors for routine care.

To test the security tools, they did something very strict: They asked everyone to give a sample of their breath (sputum) and tested it with a super-sensitive machine called "Xpert Ultra." This machine is the "Gold Standard"—it's the truth-teller. If the machine says "Intruder Found," then the intruder is there.

They then compared this truth against two common security checks:

1. The "Symptom" Check: Asking, "Do you have a cough, fever, night sweats, or weight loss?"
2. The "X-Ray" Check: Taking a picture of the lungs to see if anything looks suspicious.

The Results: The Tools Missed the Mark

1. The "Symptom" Check was like a sleepy guard.
The researchers asked, "Does anyone have a cough?"

• The Result: Only 2 out of the 9 people who actually had the intruder were coughing.
• The Analogy: Imagine a guard who only stops people who are sneezing. But the intruder is sneaky and isn't sneezing. This tool missed 8 out of 9 intruders! It was very good at not stopping innocent people (high specificity), but it was terrible at actually finding the bad guys (low sensitivity).
• Even when they expanded the question to include "Do you have a fever or lost weight?", the tool still missed most of the intruders.

2. The "X-Ray" Check was better, but still imperfect.
The researchers looked at chest X-rays.

• The Result: This tool found about 5 out of the 9 intruders. That is much better than the symptom check, but it still missed nearly half of them.
• The Catch: The X-ray is like a metal detector that beeps for any metal. Sometimes, it beeps for things that aren't the intruder. For example, if a traveler had a past fight (previous TB) that left a scar on their lungs, the X-ray would beep, thinking it found the intruder again. This caused "false alarms."

3. The "Scar" Problem (Previous TB)
This was a major finding. If a person had TB in the past, their lungs often have scars.

• The Analogy: Imagine the X-ray is a detective looking for a specific type of graffiti. If a wall was painted over years ago (past TB), the paint might still look a bit like the graffiti. The detective gets confused and thinks, "That's the graffiti!" when it's actually just an old scar.
• The Data: In people with a history of TB, the X-ray was much more likely to give a false alarm compared to people who never had TB.

4. The "Sugar" Factor (Diabetes Control)
The researchers wondered if people with poorly controlled diabetes (high blood sugar) would show the intruder more easily.

• The Result: They saw a hint that X-rays might work slightly better for people with high blood sugar, but because there were so few intruders in the study, they couldn't say for sure. It's like trying to guess the weather by looking at one cloud; you can't be certain yet.

The Bottom Line

The study concludes that for people with diabetes:

• Asking about symptoms is a weak tool. It misses too many cases.
• X-rays are better, but they aren't perfect. They still miss about half the cases, and they get confused by old scars from previous TB infections.
• The "Gold Standard" (the machine test) is still necessary because the other tools aren't reliable enough on their own.

The authors suggest that while X-rays are the most efficient tool we have right now, we need to be careful because they can cause false alarms in people with old lung scars. They also note that we need to find new, better tools to catch these "sneaky" intruders in the future, but for now, the current methods have significant blind spots.

Technical Summary

1. Problem Statement

Tuberculosis (TB) and diabetes mellitus (DM) represent a converging public health crisis, particularly in low- and middle-income countries (LMICs) like South Africa. Diabetes increases the risk of developing active TB by two- to threefold and is associated with poorer treatment outcomes. While the World Health Organization (WHO) recommends systematic TB screening for people with diabetes, the accuracy of current screening tools (symptom-based algorithms and chest X-rays) in this specific population is poorly understood.

• The Gap: Existing data largely derives from the general population. Clinical manifestations of TB in diabetics may differ (e.g., atypical chest X-ray findings, higher prevalence of cavitary lesions), and comorbidities associated with diabetes and aging may confound screening results.
• The Need: There is a critical need to validate the sensitivity and specificity of recommended screening tools specifically within a diabetic cohort to optimize detection strategies.

2. Methodology

• Study Design: A prospective cohort study utilizing baseline data from a routine primary care setting in Khayelitsha, Cape Town, South Africa.
• Population: 673 adults (≥15 years) with documented diabetes attending outpatient clinics.
  • Inclusion: Diagnosed diabetes (HbA1c ≥6.5% or fasting glucose ≥7.0 mmol/L) and current medication use.
  • Exclusion: None explicitly stated other than inability to produce sputum for the final analysis.
  • Demographics: 63.8% female; median age 53 years; 17.4% HIV-positive; 83.4% with poor glycemic control (HbA1c ≥7%).
• Intervention & Screening:
  • Symptom Screening: Assessed for cough (any duration), fever, night sweats, and weight loss.
  • Radiography: All participants underwent chest X-ray (CXR). Images were read by two physicians (consensus) and via Computer-Aided Detection (CAD4TB) with a threshold ≥60.
  • Reference Standard: A single sputum sample from every participant (regardless of symptoms or CXR findings) was tested using Xpert MTB/RIF Ultra. A positive result (including "trace" results in the primary analysis) defined prevalent TB.
• Statistical Analysis:
  • Calculated sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) with 95% confidence intervals (CI).
  • Stratified analysis by clinical factors: HIV status, prior TB history, sex, and glycemic control (HbA1c <7% vs. ≥7%).
  • Evaluated parallel screening strategies (Symptoms OR CXR).

3. Key Results

• TB Prevalence: 9 out of 661 analyzable participants (1.33%) were diagnosed with TB. Notably, all TB cases were HIV-negative.
• Symptom-Based Screening:
  • Sensitivity: Extremely low at 22.2% (95% CI: 2.1–60.0%). Only 2 of the 9 TB cases reported any cough.
  • Specificity: High at 97.5% (95% CI: 96.0–98.6%).
  • Expanding the definition to include fever, weight loss, or night sweats did not improve sensitivity and slightly reduced specificity.
• Chest X-ray (CXR) Screening:
  • Sensitivity: Improved to 55.6% (95% CI: 21.2–86.3%) for abnormalities suggestive of TB.
  • Specificity: 95.4% (95% CI: 93.4–97.0%).
  • CAD4TB: Showed higher specificity (97.7%) but lower sensitivity (37.5%) compared to human reading.
• Impact of Prior TB History:
  • Participants with a history of TB had significantly lower specificity for CXR-based tools compared to those without.
  • Specificity for CXR suggestive of TB dropped to 87.6% in those with prior TB vs. 97.2% in those without (p < 0.01). This indicates that residual scarring from previous TB leads to false positives.
• Other Factors:
  • Glycemic Control: Sensitivity appeared higher in those with poor control (HbA1c ≥7%), though confidence intervals were wide.
  • HIV Status: Specificity was lower in HIV-positive individuals for "any lung abnormality," but this difference was not significant after adjusting for prior TB history.
  • Combination Strategy: Combining symptoms and CXR did not significantly increase sensitivity (due to the low number of symptomatic cases) but slightly reduced specificity.

4. Key Contributions

• First Systematic Evaluation: This is one of the few studies to evaluate TB screening tools in people with diabetes using a rigorous design where all participants provided sputum, regardless of symptoms. This avoids the verification bias common in previous studies where only symptomatic individuals were tested.
• Quantification of Screening Failure: The study definitively demonstrates that symptom-based screening is inadequate for people with diabetes, missing nearly 80% of prevalent cases.
• Specificity Nuance: It highlights that while CXR is more sensitive than symptoms, its utility is compromised in populations with high rates of prior TB due to reduced specificity (false positives from scarring).
• CAD4TB Performance: Provides early data on the performance of AI-assisted reading (CAD4TB) in a diabetic cohort, showing high specificity but variable sensitivity.

5. Significance and Implications

• Clinical Practice: Relying solely on symptom screening for people with diabetes in high-burden settings is insufficient and likely to miss the majority of active TB cases.
• Policy Recommendations:
  • Chest X-ray should be the primary screening tool for people with diabetes, despite its suboptimal sensitivity, as it offers the highest Positive Predictive Value (PPV).
  • Contextual Interpretation: Clinicians must interpret CXR results carefully in patients with a history of TB, as residual lesions may mimic active disease, potentially leading to unnecessary confirmatory testing.
  • Future Tools: The study suggests a need for additional biomarkers (e.g., C-reactive protein) or improved AI thresholds to balance sensitivity and specificity in this dual-epidemic population.
• Limitations: The study was limited by a small number of TB cases (n=9), resulting in wide confidence intervals for sensitivity estimates. Additionally, the reference standard (single sputum Xpert) may have missed extrapulmonary TB or cases with low bacillary load.

Conclusion: The study concludes that while chest X-ray is superior to symptom screening for people with diabetes, current tools remain imperfect. The high prevalence of TB in this cohort (1.4%) underscores the urgent need for refined, feasible, and cost-effective screening strategies to control the TB-diabetes syndemic.