Do standard model assumptions realistically represent HIV dynamics in sex workers? A modelling analysis of South African data

This study demonstrates that incorporating time-varying assumptions regarding female sex workers' demographics and transmission risks into HIV models yields significantly more accurate estimates of infection dynamics and program impact in South Africa compared to traditional models relying on static assumptions.

The Gist

Imagine you are trying to predict the weather in a specific city for the next 50 years. To do this, you build a super-complex computer model. But here's the catch: your model assumes that the city's population never changes, the wind always blows at the same speed, and the temperature never shifts.

In reality, cities grow, winds change, and seasons shift. If your model ignores these changes, your weather forecast will be wrong.

This paper is about doing exactly that, but instead of weather, it's about HIV in South Africa, and instead of a city, it's focusing on female sex workers (FSW).

The Problem: The "Stuck in Time" Model

For years, scientists have used mathematical models to understand how HIV spreads. These models are like the weather forecast mentioned above. They help governments decide where to send money and medicine.

However, many of these models made a big, simple mistake: they assumed that sex workers and the risk of catching HIV never change.

• They assumed a sex worker is always the same age.
• They assumed she stays in the job for the exact same number of years.
• They assumed the chance of catching HIV from a client is always the same, no matter what year it is.

The authors of this paper thought: "Wait a minute. People age. Jobs change. And the world changes. What if we update our model to reflect real life?"

The Experiment: Testing Six Different "Realities"

The researchers used a famous South African HIV model called Thembisa. They created six different versions of the future to see which one matched reality best.

Think of these six versions as six different "simulations" or "alternate universes":

1. The "Old Way" (Constant): Everything stays exactly the same forever. (The traditional model).
2. The "Aging" Way: Sex workers get older over time, and they stay in the job longer.
3. The "Changing Risk" Way: The danger of catching HIV changes over time (specifically, it was much higher in the early days of the epidemic and has dropped since).
4. The "Realistic" Way: A mix of the above. Sex workers age, stay longer, and the risk of infection changes over time.

They ran all six simulations and then compared the results to real data collected from a massive survey of sex workers in South Africa in 2019.

The Big Reveal: The "Realistic" Way Wins

The results were surprising and very important:

• The "Old Way" (Constant) was wrong. It predicted that HIV was spreading faster than it actually is. It also underestimated how many sex workers were successfully treating their HIV (viral suppression).
• The "Realistic" Way (Time-Varying) was right. When the model allowed for sex workers to get older, stay in the job longer, and for the risk of infection to drop over time, the numbers matched the real-world survey perfectly.

Here is the analogy:
Imagine you are watching a movie where the main character is running a race.

• The Old Model is like a movie where the runner never gets tired, never slows down, and the track never changes. It predicts they will finish in 10 minutes.
• The Real World is like a movie where the runner gets older, the track gets muddy, and they slow down. They actually finish in 20 minutes.
• If you use the "Old Model" to plan a race, you'll be shocked when the runner takes 20 minutes. You might think, "Wow, the race is harder than we thought!" or "We need more runners!" when actually, the runner just needs a different strategy.

Why Does This Matter? (The "So What?")

This isn't just about math; it's about money and lives.

1. Overestimating the Problem: The "Old Model" suggested that sex work was responsible for 20% of all new HIV infections in the future. The "Realistic Model" says it's actually only about 9-13%.

   • Why this matters: If you think sex work is causing 20% of the problem, you might panic and throw all your money at it. If it's actually 10%, you might be wasting resources that could be used elsewhere, or you might be misjudging how well current programs are working.

2. Underestimating Success: The "Old Model" thought sex workers were getting infected faster and treating HIV less effectively. The "Realistic Model" showed that programs are actually working better than we thought! More sex workers are staying in the job longer, getting older, and successfully suppressing the virus.

3. Better Planning: By using the "Realistic Model," governments can stop guessing and start making decisions based on what is actually happening. They can see that while sex workers are still a high-risk group, the epidemic is changing, and their strategies need to change with it.

The Bottom Line

This paper is a wake-up call for scientists and policymakers. You cannot use yesterday's assumptions to predict tomorrow's reality.

If you want to fight HIV effectively, you have to acknowledge that people age, behaviors change, and risks evolve. By updating the "weather forecast" to include these real-life changes, we can finally see the storm clearly and build the best possible shelter for everyone.

Technical Summary

Technical Summary: Do standard model assumptions realistically represent HIV dynamics in sex workers?

1. Problem Statement
Mathematical models are critical tools for understanding HIV transmission dynamics and guiding prevention programs for key populations, specifically Female Sex Workers (FSW) in sub-Saharan Africa. However, standard models (such as the widely used Thembisa model) often rely on time-invariant assumptions, positing that FSW characteristics (age, duration of sex work) and biological transmission risks (per-act client-to-FSW transmission probability) remain constant over the course of an epidemic.

Empirical evidence suggests these assumptions are unrealistic:

• Demographics: The average age of FSW and the duration of their engagement in sex work have been increasing in South Africa.
• Transmission Risk: Evidence from Kenya and other settings suggests that per-act transmission probabilities were likely higher during the early epidemic and have declined over time due to factors such as reduced STI prevalence, behavioral changes, or the depletion of highly susceptible individuals.

The paper argues that failing to account for these temporal variations may bias model estimates, leading to inaccurate projections of HIV incidence, prevalence, and the population attributable fraction (PAF) of sex work to the overall epidemic.

2. Methodology
The study utilized Thembisa v4.7, a deterministic compartmental model of the HIV epidemic in South Africa, to simulate FSW dynamics under six distinct scenarios.

• Scenarios: The authors combined two specifications for FSW characteristics with three specifications for transmission risk:

  • FSW Characteristics:
    • (a) Time-invariant: Constant mean age (29 years) and duration (3 years).
    • (b) Time-varying: Mean age increased from 26.4 to 32.2 years; mean duration increased from 2.7 to 7.4 years (1996–2019).
  • Transmission Risk (Client-to-FSW):
    • (1) Constant: Fixed per-act probability.
    • (2) Exponential early-epidemic decline: Risk starts high and decays exponentially over calendar time.
    • (3) Exposure-dependent decline: Risk decays as a function of cumulative exposure to HIV-positive clients (incorporating client prevalence, ART coverage, and duration).
  • Resulting Scenarios: 1a, 1b, 2a, 2b, 3a, 3b.

• Calibration: Each scenario was calibrated using a Bayesian approach against historical HIV prevalence data from:

  • Antenatal clinic (ANC) surveys (1990–present).
  • National household surveys (2005–2022).
  • FSW-specific surveys (late 1990s–2019).
  • Note: Scenarios 2a–3b included an additional parameter for early-epidemic excess risk, requiring 11 calibrated parameters vs. 10 for scenarios 1a/1b.

• Validation: Model outputs (incidence, prevalence, viral load suppression) were validated against external data not used in calibration:

  • The 2019 South African National FSW Survey (incidence, prevalence, viral suppression).
  • Regional meta-analysis estimates of Incidence Rate Ratios (IRRs) for Eastern and Southern Africa.

3. Key Contributions

• Systematic Sensitivity Analysis: This is the first study to systematically quantify how time-invariant vs. time-varying assumptions regarding FSW demographics and transmission biology affect HIV model outputs in a generalized epidemic setting.
• External Validation: The study moves beyond fitting calibration data to testing model performance against independent, high-quality survey data (2019 FSW survey) and regional benchmarks.
• Re-evaluation of Attributable Risk: The study challenges existing estimates of the Population Attributable Fraction (PAF) of sex work, demonstrating that standard assumptions may significantly overstate the contribution of sex work to new infections.

4. Key Results

• Model Fit and Validation:

  • All scenarios fit historical prevalence data well.
  • Best Fit: Scenarios 2b (Increasing FSW age/duration + Exponential decline in risk) and 3b (Increasing FSW age/duration + Exposure-dependent decline in risk) showed the closest agreement with the 2019 external validation data.
    • Incidence: ~5.0–5.4 per 100 person-years (vs. observed ~4.6).
    • Prevalence: ~61–62% (vs. observed 62.1%).
    • Viral Suppression: ~60% (vs. observed 58.9%).
  • Poor Fit: Scenarios assuming constant characteristics and constant risk (1a, 1b) overestimated HIV incidence and underestimated viral suppression and prevalence.

• Incidence Rate Ratios (IRRs):

  • Time-varying scenarios produced IRRs (FSW incidence vs. general female incidence) that were more consistent with regional meta-analyses.
  • Constant-risk scenarios implied a faster decline in FSW incidence relative to the general population than observed in regional data.

• Population Attributable Fraction (PAF):

  • Significant Discrepancy: Time-invariant assumptions (1a/1b) attributed >20% of new HIV infections in 2025 to commercial sex work.
  • Time-Varying Reality: Scenarios 2b and 3b attributed only 9–13% of new infections to sex work.
  • Long-term Projections: Under time-varying assumptions, the PAF stabilizes or grows slowly. Under constant assumptions, the PAF is projected to rise sharply, potentially misleading resource allocation.

5. Significance and Implications

• Policy Impact: The findings suggest that relying on standard, time-invariant assumptions leads to an overestimation of the role of sex work in sustaining the epidemic. This could result in misdirected resources or an overly pessimistic view of the effectiveness of existing programs.
• Program Planning: Incorporating time-varying parameters (aging FSW, longer duration, declining transmission risk) yields estimates that align better with empirical reality. This supports more accurate planning for FSW-specific interventions.
• Future Modeling: The study calls for a critical reassessment of assumptions in HIV modeling globally. As epidemics mature and key populations age, static models may fail to capture the shifting dynamics of transmission, particularly regarding the "concentration" of the epidemic in key populations.
• Funding Context: Given recent cuts in funding for key populations, accurate modeling is essential to demonstrate the continued need for targeted interventions without inflating the perceived contribution of these groups to the general epidemic.

In conclusion, the paper demonstrates that HIV model estimates for sex workers are highly sensitive to underlying assumptions. Moving toward time-varying, empirically grounded parameters improves the reliability of model outputs and ensures that public health strategies are based on realistic epidemic dynamics.