📊 epidemiology

Causal effect of loss to follow-up on mortality in a population-based tuberculosis cohort in Brazil

Using a sequential landmark cohort analysis of over 171,000 tuberculosis patients in Brazil, this study demonstrates that loss to follow-up at any point during treatment substantially increases late TB-attributable mortality, highlighting the critical need to prevent treatment interruptions as a programmatic priority.

Original authors: Lepka de Lima, E., Lindoso, A. A., Orlandi, G., Fukasava, S., Martinez, C., Croda, J., Horsburgh, C. R., Ranzani, O., Brooks, M. B., Andrews, J. R.

Published 2026-07-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Lepka de Lima, E., Lindoso, A. A., Orlandi, G., Fukasava, S., Martinez, C., Croda, J., Horsburgh, C. R., Ranzani, O., Brooks, M. B., Andrews, J. R.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are trying to climb a very long, steep mountain called "Recovery." You have a guide (the doctor) and a map (the treatment plan) that tells you exactly how to get to the top. But, for many people, the journey gets interrupted. They stop climbing, leave the path, and disappear into the fog. This is what researchers call Loss to Follow-Up (LTFU).

For a long time, scientists have known that people who stop their tuberculosis (TB) treatment often end up getting sick or dying. But they struggled to answer a specific question: Is the death caused by the fact that they stopped the medicine, or were they already so sick or vulnerable that they would have died anyway, regardless of whether they stayed on the path?

It's like trying to figure out if a car crash was caused by the driver hitting the brakes too late, or if the car was already broken before they even started driving. The problem is that the "drivers" who stop treatment are a mixed bag: some are young and healthy but just distracted, while others are very sick and struggling to stay alive.

The "Time Travel" Problem

The researchers in this study faced a tricky math problem called "Immortal Time Bias."

Imagine you are timing a race. If you only start the clock for the "Lost" group after they have successfully walked for 30 days without falling, you are cheating. You are giving them a "free pass" on those first 30 days. They had to survive those 30 days just to be counted in your study. Meanwhile, the "Stayed" group is being timed from the very first step. This makes the "Lost" group look artificially healthier and safer than they really are, at least in the short term.

To fix this, the researchers used a clever method called a "Sequential Landmark" analysis. Think of it like setting up a series of checkpoints every single month.

Checkpoint 1: We look at everyone who was still climbing at Month 1. We compare those who stopped right then with those who kept going.
Checkpoint 2: We do the same for Month 2, Month 3, and so on.

By aligning the start times perfectly, they removed the "free pass" and got a true picture of what happens when someone stops.

What They Found

The study looked at over 170,000 people with TB in São Paulo, Brazil, over ten years. Here is what the "time-travel" math revealed:

The "Late" Danger: In the first few months after stopping, the data looked confusing (because of the bias mentioned above). But once they looked at the long-term (6 months to 2 years after stopping), the picture became very clear. Stopping treatment doubled or even tripled the risk of dying compared to those who kept going.
The Cause of Death: This is the most important part. The extra deaths were almost entirely TB-related.
- Think of it like this: If you stop taking your antibiotics, the TB bacteria (the enemy) wakes up and attacks. The patients didn't die from car accidents or old age (non-TB causes); they died because the enemy they were fighting won.
- This proves that stopping the treatment itself is the direct cause of the death, not just the fact that the person was already poor or sick.
Who Suffers Most?
- Relative Risk: Younger, healthy people who stopped treatment saw the biggest percentage increase in risk. It's like a small pebble causing a huge landslide on a dry hill.
- Absolute Risk: However, the people who actually died in the highest numbers were those who were already very vulnerable, specifically people living with HIV. For them, stopping treatment was like removing the last safety net; the risk of death jumped by a massive amount (3 percentage points higher).

The "Return" Trap

The study also looked at people who stopped, got lost, and then came back to the clinic later.

The Reality: People who returned to treatment had much higher death rates than those who never left.
The Metaphor: Imagine a hiker who falls off the trail, wanders in the woods for months, and then climbs back up. By the time they get back to the path, they are exhausted, injured, and the weather has turned. Even if they finish the climb, they are in much more danger than the hiker who never left the trail in the first place.

The Bottom Line

This study acts like a high-definition camera that finally focused on the blurry picture of "stopping treatment." It confirms that stopping TB treatment is a direct, causal path to death, specifically from TB.

The researchers conclude that we can't just tell people to "take their pills." We have to fix the things that make them stop in the first place—like homelessness, addiction, and lack of support. If we don't keep people on the path, the mountain becomes much more deadly, not because the mountain changed, but because they left the safety of the guide.

Technical Summary: Causal Effect of Loss to Follow-up on Mortality in a Brazilian Tuberculosis Cohort

Problem Statement
Loss to follow-up (LTFU) during tuberculosis (TB) treatment represents a critical programmatic gap, contributing to drug resistance, ongoing transmission, and increased mortality. However, quantifying the direct causal effect of LTFU on mortality has been historically difficult due to significant methodological challenges. Specifically, standard observational comparisons are confounded by "confounding by indication" (where factors driving disengagement also drive poor outcomes) and distinct forms of "immortal-time bias."

Immortal-time bias arises because:

Survival Requirement: To be classified as LTFU, a patient must survive long enough to disengage from care (typically defined as 30 days of absence). Patients who die immediately after treatment initiation are classified as "on-treatment deaths" and can never enter the LTFU group, making LTFU appear spuriously protective in early follow-up.
Completion Bias: Conversely, comparing LTFU patients to those who complete treatment introduces bias because treatment completion is a post-baseline outcome attainable only by surviving the full course.

Methodology
The authors conducted a retrospective cohort study using linked data from the São Paulo State TB surveillance system (TBweb) and the national mortality registry (SIM) for the period 2013–2023. The study included 171,048 individuals aged ≥15 initiating their first TB treatment.

To address the aforementioned biases and estimate the causal effect, the study employed a sequential landmark cohort analysis:

Landmark Design: The authors constructed monthly cohorts corresponding to each month of potential treatment interruption ( $m = 1$ to $6$).
Symmetric Grace Period: To align eligibility and address the programmatic definition of LTFU (which requires 30 days of absence), a symmetric 30-day grace period was applied. Time at risk began 30 days after the interruption month ( $m + 30$ days) for both the LTFU group and the continued-treatment group. This ensured both groups had survived the same duration to be eligible for the comparison.
Statistical Modeling:
- Hazard Ratios: Adjusted Cox proportional hazards models were used to estimate hazard ratios (aHR) for all-cause and cause-specific mortality (TB-attributable vs. non-TB) in "late" windows (6–24 months post-landmark).
- Negative Control: Non-TB mortality served as a within-cohort negative control to distinguish the causal effect of interrupted therapy from residual confounding by general vulnerability.
- Absolute Risk: Standardized (g-formula) absolute mortality risk differences were computed to quantify the burden on the population scale.
- Sensitivity Analyses: Analyses included multiple imputation for missing data, alternative LTFU timing definitions, and restriction to deaths classified solely by ICD-10 codes in the mortality registry.

Key Results

Cohort Characteristics: Of the 171,048 individuals, 20,830 (12.2%) experienced LTFU. LTFU patients were significantly younger, more frequently male, and had higher prevalence of homelessness, substance use, and HIV co-infection compared to those who remained in care.
Causal Effect on Mortality:
- Late Mortality: LTFU at any point in therapy substantially increased late mortality (6–24 months). Adjusted hazard ratios ranged from 1.83 (95% CI 1.27–2.64) for LTFU at month 6 to 2.85 (95% CI 2.34–3.48) for LTFU at month 3.
- Absolute Risk: The standardized late-window mortality risk difference reached up to 1.6 percentage points.
- Early Mortality: In the early window (0–6 months), hazard ratios were often below unity (e.g., 0.63 at month 1), reflecting the residual survival selection bias inherent in the LTFU definition rather than a protective effect of disengagement.
Cause-Specific Evidence: The excess mortality was concentrated in TB-attributable deaths (aHR range 1.60–4.36). In contrast, the effect on non-TB mortality was essentially null (aHR range 0.96–1.48). This divergence supports a causal pathway through interrupted therapy rather than generalized social vulnerability.
Subgroup Analysis:
- Relative Effects: The relative hazard (aHR) was highest in younger (15–24 years) and stably housed individuals, groups with lower baseline mortality.
- Absolute Burden: Despite lower relative hazards, the absolute excess mortality (risk difference) was largest among people living with HIV (3.0 percentage points) and homeless individuals, reflecting their higher baseline risk.
Return to Care: Patients who returned to care after LTFU had significantly higher mortality (9.9% at 24 months) compared to matched patients who remained lost but alive (1.6%), and higher than the counterfactual estimate if they had completed treatment (3.5%).

Significance and Claims
The paper claims to be the first analysis to estimate the causal effect of LTFU on mortality using a sequential landmark cohort analysis with symmetric grace-period eligibility. Its primary significance lies in:

Methodological Correction: Successfully addressing the immortal-time bias that has previously rendered standard observational comparisons of LTFU and treatment continuation difficult to interpret.
Causal Inference: Providing robust evidence that LTFU itself, beyond underlying social and clinical vulnerabilities, carries a measurable, TB-specific mortality cost. The concentration of excess deaths in TB-attributable causes, rather than non-TB causes, confirms that the mortality risk is driven by the interruption of therapy.
Programmatic Implications: The findings suggest that preventing LTFU is a critical priority. The authors argue that adherence-focused approaches alone are insufficient; interventions must address the structural drivers of disengagement (housing instability, substance use, HIV co-infection) to reduce both the incidence of LTFU and the resulting mortality burden.

The authors remain modest regarding limitations, acknowledging that residual confounding from unmeasured disease severity markers (e.g., viral load, CD4 count) is possible, and that mortality ascertainment relied partly on probabilistic linkage which may under-ascertain deaths in highly mobile populations. However, the cause-specific contrast and sensitivity analyses support the robustness of the primary findings.

The "Time Travel" Problem

What They Found

The "Return" Trap

The Bottom Line

Technical Summary: Causal Effect of Loss to Follow-up on Mortality in a Brazilian Tuberculosis Cohort

More like this