Time to registry discontinuity in Tanzania's national HIV care registry: a survival analysis of population mobility patterns

This study of over 2 million HIV patients in Tanzania reveals that early registry discontinuity is common and geographically linked to population mobility, suggesting that many apparent losses to follow-up actually reflect administrative gaps in tracking mobile patients across facilities rather than true disengagement from care.

The Gist

Imagine you are running a massive library where millions of people have signed up to borrow books. The library's main goal is to make sure everyone keeps coming back to check out their next book on time.

In the world of HIV care, Tanzania has a similar "library" called a National Registry. It's a giant digital list that tracks millions of people who start taking life-saving medication. Health officials use this list to count how many people "drop out" (stop coming to the clinic). This is called Loss to Follow-Up (LTFU).

For years, officials assumed that if a person's name disappeared from the list, it meant they had stopped taking their medicine, perhaps because they were too sick, felt ashamed, or couldn't afford the trip to the clinic.

But this study suggests the library might be counting "missing" people who are actually just borrowing books from a different branch.

Here is the story of what the researchers found, broken down simply:

1. The "Silent Transfer" Mystery

Imagine you live in a small town (a rural district). You go to the local library to get your first book. A few weeks later, you move to a big city for a new job. You go to the city library to get your next book.

To the small town library, you look like a "dropout." They see you signed up, but you never came back. They mark you as "Lost."
To the city library, you are a "new customer." They see you signing up for the first time.

Because the two libraries don't talk to each other, the system thinks:

• The small town lost a patient.
• The city gained a new patient.
• Reality: You never stopped reading; you just changed locations.

In Tanzania, this "Silent Transfer" is happening on a massive scale. People move for jobs (mining), for grazing animals (pastoralists), or for trade (border towns). When they move, they often start treatment at a new clinic, but the old clinic doesn't know.

2. The "Front-Loaded" Drop

The researchers looked at the data like a movie playing in fast forward. They noticed something strange about when people disappeared from the lists.

If people were quitting because they were sick or gave up, you would expect them to leave slowly over time, like water dripping from a leaky faucet.

Instead, the data showed a massive splash right at the beginning.

• 9.6% of people vanished within the first month.
• 17.8% vanished within six months.

It was like a crowd of people running out the door immediately after walking in. This "front-loaded" pattern suggested that these weren't people giving up on treatment; they were people who moved away almost immediately after registering.

3. The Geography of Movement

The study mapped these "dropouts" across Tanzania and found a clear pattern, like a weather map showing where the wind blows strongest.

• Stable Areas: In quiet, rural districts where people mostly stay put, the "dropout" rate was lower.
• Mobile Areas: In places known for movement—border towns, mining camps, cities with lots of migrants, and areas where herders move their cattle—the dropout rate was much higher.

It's as if the "dropout" rate was a shadow cast by the movement of people. Wherever people moved, the registry lost them.

4. The "New" Patient Illusion

The study also noticed a huge spike in "new" registrations in 2017. Why? Because that's when the country updated its computer system and merged old lists.

Many of these "new" patients weren't actually starting treatment for the first time in their lives. They were people who had been treated in a different district years ago, and now their names were being added to the national list for the first time. This made it look like thousands of new people were starting, but many were just being "re-registered" because they moved or the system changed.

The Big Takeaway

The main lesson from this paper is that Tanzania's "dropout" numbers might be lying.

When the system says, "We lost 17% of our patients in six months," it might actually mean, "We lost track of 17% of our patients because they moved to a different town, and our two town libraries don't talk to each other."

Why does this matter?

• False Alarm: Health officials might think their clinics are failing or that patients are quitting, when in reality, the patients are just continuing their treatment elsewhere.
• Wasted Effort: Resources might be spent trying to find people who aren't actually missing, just relocated.
• The Fix: The study argues that Tanzania needs a "Universal Library Card" system. Instead of tracking patients by which building they visit, they need a system that tracks the person across the whole country. If a patient moves from a village to a city, the system should automatically update their file so they don't get marked as "Lost."

In short: The patients aren't disappearing; the tracking system is just too narrow to see them moving.

Technical Summary

1. Problem Statement

Routine HIV program monitoring relies heavily on Early Loss to Follow-Up (LTFU) indicators (e.g., absence of a clinic visit within 30–90 days of enrollment) to assess program performance and patient retention. However, these indicators operate on a critical assumption: that patients remain geographically stable and that absence from a specific facility implies disengagement from care.

In Tanzania, as in many sub-Saharan African nations, internal population mobility is substantial due to pastoralism, mining, cross-border trade, and rural-to-urban migration. Current national registries track patients within specific administrative districts but do not link records across facility or district boundaries. Consequently, patients who relocate and continue treatment at a new facility appear as "lost" in their original registration district. This phenomenon, known as "silent transfer," leads to the systematic misclassification of mobile patients as LTFU, potentially underestimating true treatment retention and distorting geographic performance metrics.

2. Methodology

• Study Design: A national, registry-based observational analysis using routinely collected data.
• Data Source: The HIV care registry maintained by the National AIDS and Sexually Transmitted Infections Control Programme (NASHCoP).
• Cohort: 2,136,207 individuals with recorded district registration and visit dates between January 1, 2017, and December 31, 2021.
• Outcome Definition:
  • Registry Discontinuity: Defined as the time interval between the first recorded visit and the absence of further recorded visits within the registering district.
  • Thresholds: Discontinuity was measured at 30, 60, 90, and 180 days post-registration.
  • Censoring: To avoid bias from the end of the observation window, threshold-specific eligibility cohorts were constructed (e.g., only individuals registering before July 2021 were included in the 180-day analysis).
• Geographic Classification: Districts were categorized into five population mobility corridors:
  1. Stable districts (comparison group).
  2. Urban migration districts.
  3. Mining areas.
  4. Pastoralist regions.
  5. Border districts.
• Statistical Analysis:
  • Kaplan-Meier Survival Analysis: Used to estimate time-to-discontinuity curves.
  • Log-Rank Test: Used to compare survival curves across the five mobility corridors.
  • Software: R (version 4.5.2).

3. Key Contributions

• Reframing LTFU: The study challenges the interpretation of early LTFU as a direct measure of patient disengagement, proposing instead that it is often a proxy for administrative discontinuity caused by geographic mobility.
• Geographic Structuring: It demonstrates that early registry disappearance is not random but follows a distinct, predictable geographic structure aligned with known population mobility patterns.
• Registry Artifact Identification: The analysis highlights that the "first recorded visit" in the national registry often reflects administrative enrollment or database integration rather than the true initiation of lifetime HIV care, evidenced by a massive spike in new registrations in 2017.

4. Key Results

• High Early Discontinuity Rates:
  • 30 days: 9.6% of patients had no further recorded visits.
  • 90 days: 13.5% discontinued.
  • 180 days: 17.8% discontinued.
• Temporal Pattern: The Kaplan-Meier survival curve showed a steep early decline followed by a prolonged plateau. This indicates that the vast majority of "losses" occur immediately after registration, rather than accumulating gradually over time (which would suggest behavioral disengagement).
• Mobility Corridor Differences:
  • There was a statistically significant difference in time-to-discontinuity across corridors (Log-rank $\chi^2$ = 15,372, p < 0.001).
  • Highest Discontinuity: Border districts (13.6% at 30 days; 27.4% at 180 days), followed by urban migration and mining zones.
  • Lowest Discontinuity: Stable districts (9.1% at 30 days; 20.0% at 180 days).
• Geographic Clustering: High discontinuity rates were concentrated in specific regions (e.g., Ngorongoro, Mpanda, Katavi, Geita) known for high mobility, while stable regions (e.g., Iringa, Songwe) showed significantly lower rates.
• Enrollment Anomaly: Nearly one million individuals were recorded as "newly registered" in 2017, suggesting the registry was integrating historical data rather than capturing only new clinical initiations.

5. Significance and Implications

• Indicator Validity: The study concludes that facility-based retention indicators systematically underestimate treatment continuity in mobile populations. High LTFU rates in border, mining, and pastoralist regions may reflect successful patient movement rather than program failure.
• Policy Recommendations:
  • System Integration: There is an urgent need for monitoring systems capable of linking patient records across facilities and administrative boundaries (e.g., via unique patient identifiers or biometric systems).
  • Metric Adjustment: Program managers must interpret early LTFU signals within the context of local mobility patterns to avoid misallocating resources or penalizing districts with high mobility.
  • Silent Transfer Recognition: The findings validate the existence of "silent transfers," where patients are lost to the registry in one district but remain in care in another, necessitating a shift from facility-centric to person-centric surveillance.

In summary, the paper provides robust evidence that a substantial proportion of early LTFU in Tanzania is an artifact of the monitoring system's inability to track mobile patients, rather than a failure of patients to adhere to treatment.