📄 medicine

A Clinical Risk Score for Predicting Anti-Tuberculosis Drug-Induced Liver Injury During Standard Treatment for Drug-Susceptible Tuberculosis

This retrospective cohort study developed and validated a five-variable clinical risk score incorporating age, co-administered hepatotoxic drugs, hypoalbuminemia, aspartate transaminase levels, and NAT2 slow acetylator phenotype to accurately predict anti-tuberculosis drug-induced liver injury in patients undergoing standard treatment for drug-susceptible tuberculosis.

Original authors: Thanyanuch Sanchat, Charoen Chuchottaworn, Prapasri Kulalert

Published 2026-08-31

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Thanyanuch Sanchat, Charoen Chuchottaworn, Prapasri Kulalert

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Tuberculosis remains a formidable public health challenge, particularly in Southeast Asia, where the standard treatment involves a six-month course of powerful antibiotics. While these drugs are highly effective at killing the bacteria, they carry a significant risk: they can damage the liver. This condition, known as drug-induced liver injury, is a common and serious side effect that can force doctors to stop treatment, switch medications, or leave patients vulnerable to the disease returning. The risk of this liver damage is not random; it is influenced by a mix of factors including a patient's age, their existing health conditions, and their unique genetic makeup. One specific gene, which controls how the body processes one of the main tuberculosis drugs, plays a particularly large role. Some people's bodies break down the medication slowly, causing it to build up to toxic levels, while others process it quickly. Currently, doctors lack a simple, practical way to predict which patients will face this danger before they even start taking the pills, often leading to a reactive approach where liver damage is only addressed after it has occurred.

Researchers at the Central Chest Institute of Thailand set out to change this by creating a straightforward tool to identify high-risk patients in advance. They looked back at the medical records of 426 patients treated for drug-susceptible tuberculosis between 2018 and 2024. All these patients received the standard six-month regimen, and crucially, every one of them underwent genetic testing to determine their specific liver-processing type. The team found that 111 of these patients, or about 26 percent, developed liver injury during their treatment. By comparing the medical histories and blood test results of those who got sick with those who did not, the researchers identified five key warning signs that consistently appeared in the group that suffered liver damage.

The most powerful predictor turned out to be the genetic profile. Patients with a "slow acetylator" phenotype, meaning their bodies naturally process the drug slowly, were far more likely to develop liver injury than those with faster processing speeds. Beyond genetics, the researchers found that being 60 years of age or older, taking other medications that can also stress the liver, having low levels of albumin (a protein that indicates nutritional health), and having elevated levels of a specific liver enzyme called aspartate transaminase before starting treatment were all significant risk factors. Interestingly, the study found that other factors often suspected in previous research, such as alcohol use or being female, did not show a strong link to liver injury in this specific group of Thai patients.

Using these five confirmed factors, the team built a scoring system that assigns points to each risk factor a patient possesses. A patient's total score can range from zero to 12.5, with higher numbers indicating a greater likelihood of liver damage. When the researchers tested this score, it proved to be a reliable guide. It successfully distinguished between patients who would develop liver injury and those who would not, with a performance level that suggests it is a robust tool for clinical use. The model was tested internally to ensure it was not just a fluke of the specific data used, and it held up well, showing that it could be applied to new patients with confidence.

The researchers determined that a score of 7 or higher serves as a critical threshold. Patients reaching this level have a significantly increased risk of developing liver injury, with the odds being more than eight times higher than for those with lower scores. While the tool is excellent at identifying those who are definitely at risk, it is less sensitive at catching every single potential case, meaning it is best used to flag the most vulnerable individuals for closer monitoring rather than as a blanket screen for everyone. This distinction is vital for clinical practice, as it allows doctors to focus their attention and resources on the patients who need them most.

This new scoring system offers a practical path forward for tuberculosis care in Thailand and potentially beyond. By combining a simple genetic test with routine blood work and a review of a patient's age and other medications, doctors can now predict who is likely to suffer liver damage before the first dose is even taken. For those identified as high-risk, medical teams can implement stricter monitoring schedules or consider alternative treatment strategies to protect the liver while still curing the tuberculosis. The study acknowledges that it was conducted at a single, specialized center and relied on past records, which means future research in broader settings is needed to confirm these findings. However, the development of this score represents a significant step toward personalized medicine, moving tuberculosis treatment from a one-size-fits-all approach to a safer, more tailored strategy that keeps patients on their life-saving medication without compromising their liver health.

Technical Summary: A Clinical Risk Score for Predicting Anti-Tuberculosis Drug-Induced Liver Injury (ATDILI)

Problem Statement
The standard 6-month regimen (2HRZE/4HR) for drug-susceptible tuberculosis (DS-TB) is highly effective but frequently associated with adverse drug reactions, particularly anti-tuberculosis drug-induced liver injury (ATDILI). ATDILI can compromise treatment continuity, necessitate regimen modifications, and lead to poorer outcomes. While clinical risk factors (e.g., age, comorbidities) and genetic factors (specifically NAT2 polymorphisms) are known to influence ATDILI risk, there is currently no practical, standardized clinical tool that integrates both genetic and clinical data to predict risk before treatment initiation. Existing models often rely on complex variables or post-treatment laboratory parameters, limiting their utility in routine practice, particularly in the Thai healthcare setting where NAT2 slow acetylator phenotypes are common.

Methodology
This study employed a retrospective cohort design involving patients treated at the Central Chest Institute of Thailand between January 2018 and December 2024.

Cohort Selection: From an initial pool of 2,688 patients, 849 underwent NAT2 genotyping. After applying exclusion criteria, 426 patients receiving the standard 2HRZE/4HR regimen were included in the final analysis (111 cases of ATDILI, 315 controls).
Data Collection: Baseline characteristics included demographics, comorbidities, concomitant hepatotoxic medication use, liver function tests (AST, ALT, bilirubin, albumin), hematologic markers, and NAT2 acetylator phenotypes.
Model Development:
1. Univariable Analysis: Candidate predictors were screened using logistic regression. Variables with $p < 0.20$ and clinical relevance were selected.
2. Multivariable Modeling: A multivariable logistic regression model was developed using backward elimination. Multicollinearity was assessed via Variance Inflation Factor (VIF).
3. Score Derivation: Regression coefficients were converted into a point-based risk score by dividing each coefficient by the smallest retained coefficient and rounding to the nearest 0.5.
Validation: Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AuROC) for discrimination and the Hosmer–Lemeshow test for calibration. Internal validation was performed using 800 bootstrap resamples to assess optimism and overfitting. Decision curve analysis was used to evaluate clinical utility.

Key Contributions
The primary contribution of this work is the development and internal validation of a five-variable clinical risk score specifically designed for the Thai population receiving standard DS-TB therapy.

Integration of Genetics and Clinical Data: The model uniquely incorporates NAT2 slow acetylator phenotype alongside clinical variables, addressing a gap in existing prediction tools.
Simplicity and Practicality: The resulting score utilizes only five objective, clinically available variables, making it feasible for routine pre-treatment risk stratification.
Local Contextualization: Unlike models derived from different treatment regimens (e.g., thrice-weekly) or populations, this tool is tailored to the daily 2HRZE/4HR regimen standard in Thailand.

Results

Incidence: The study observed an ATDILI incidence of 26.1% (111/426 patients).
Predictors: The final model retained five significant predictors:
1. Age $\ge$ 60 years
2. Co-administration of hepatotoxic drugs
3. Elevated baseline Aspartate Transaminase (AST) levels
4. NAT2 slow acetylator phenotype
5. Hypoalbuminemia
Model Performance:
- Discrimination: The model demonstrated good discriminatory ability with an apparent AuROC of 0.80 (95% CI: 0.75–0.85). Internal validation confirmed stability with an optimism-corrected AuROC of 0.79.
- Calibration: The model showed good agreement between predicted and observed risks (Hosmer–Lemeshow $p = 0.80$ ; calibration-in-the-large = 0.00).
- Risk Score: The total score ranges from 0 to 12.5. Each one-point increase in the score was associated with a 1.50-fold higher risk of ATDILI.
Clinical Cutoff: A cutoff score of $\ge$ 7 was identified as optimal for clinical utility. At this threshold, the model achieved a specificity of 92.13% and a Positive Predictive Value (PPV) of 63.1%, though sensitivity was lower at 41.84%. Patients scoring $\ge$ 7 had an Odds Ratio of 8.42 (95% CI: 4.72–15.02) for developing ATDILI.

Significance and Claims
The authors claim that the developed ATDILI risk score is an "accurate and practical tool" for identifying DS-TB patients at high risk for liver injury prior to treatment initiation. The study posits that this tool can support:

Risk Stratification: Enabling clinicians to identify high-risk patients (score $\ge$ 7) for closer monitoring, particularly during the first 8 weeks of treatment.
Individualized Care: Facilitating the review of concomitant hepatotoxic medications and potentially guiding individualized strategies, such as pharmacogenetic-guided isoniazid dose adjustments or alternative regimens, though the authors note these specific interventions require further evaluation.
Clinical Utility: Decision curve analysis indicated that the score provides a higher net benefit than "treat-all" or "treat-none" strategies across a range of threshold probabilities.

The authors maintain a modest stance, acknowledging limitations such as the retrospective design, single-center nature (a referral center), and incomplete data on viral hepatitis and alcohol use. They emphasize that while the tool supports pre-treatment stratification, future prospective studies with external validation are necessary to confirm generalizability and improve predictive accuracy.

Technical Summary: A Clinical Risk Score for Predicting Anti-Tuberculosis Drug-Induced Liver Injury (ATDILI)

More like this