How the Risk Engine Works

From Self Report to Event Level Probability

The Risk Engine is designed to map a finite set of structured inputs onto a quantitative estimate of HIV acquisition risk for a single exposure. It does not rely on heuristic scoring or simple checklists. Instead, it applies a Bayesian inspired event model that proceeds through a fixed four stage sequence:

• Specification of a prior probability that the partner is living with HIV, conditional on demographic and behavioural context
• Selection of an act specific per exposure transmission probability given a partner with untreated HIV infection
• Application of multiplicative modifiers that represent the effect of prevention or amplifying factors on transmission efficiency
• Combination of these components into a single calibrated percentage and corresponding “1 in X” representation

In practice, this means that each user session instantiates a small probabilistic model of one encounter. The model parameters are anchored in published estimates from bodies such as the CDC, WHO, NHS and peer reviewed meta analyses, then adjusted only within ranges supported by those data.

Users can initiate a full assessment at any time via the HIV Risk Assessment page.

1. Estimating Partner HIV Prevalence

The first stage is the construction of a prior probability that the partner was living with HIV before the encounter. This prior is derived from location and gender specific prevalence estimates based on national and regional surveillance reports.

Where epidemiological data distinguishes higher prevalence key populations, such as men who have sex with men or individuals with injecting drug use, the engine makes use of those stratified estimates rather than applying a single population average. This produces a prior distribution that is more representative of the sexual network in which the encounter occurred.

The baseline prior is then adjusted using multiplicative factors when users report partner level characteristics associated with elevated HIV likelihood, for example:

• Injecting drug use and sharing of injecting equipment
• Engagement in sex work or frequent anonymous partnering
• Known contact with high prevalence sexual networks or concurrent sexually transmitted infections

These adjustments are implemented as deterministic multipliers applied to the base prevalence value, yielding a partner specific prior that serves as the starting point for subsequent calculations.

2. Modelling Act Specific Transmission Risk

The second stage incorporates the biological risk associated with the specific sexual act and role described by the user. The engine distinguishes between multiple exposure types, including receptive anal intercourse, insertive anal intercourse, receptive vaginal intercourse, insertive vaginal intercourse and selected oral exposures.

For each act type, the model uses a per exposure transmission probability that represents the estimated probability of HIV transmission when the partner has untreated HIV and no prevention method is in place. These estimates are derived from large cohort studies and meta analytic syntheses of serodiscordant couples and observational data.

In statistical terms, these values function as likelihood components: they describe the conditional probability of infection given that the partner is HIV positive and that the exposure takes the specified form. The engine encodes these likelihoods numerically so they can be combined in a consistent way with the prior and modifier terms.

3. Incorporating Prevention and Amplifying Factors

The third stage applies multiplicative modifiers that capture the effect of real world conditions that either attenuate or amplify the baseline act specific risk. These modifiers operate on the per exposure probability rather than on the prior.

• Condom use, including integrity throughout the exposure and any reported break, slip or removal events
• Use of oral pre exposure prophylaxis (PrEP) by the user, with assumptions about adherence consistent with guideline summaries
• Use of antiretroviral therapy by the partner and, where information is available, reported viral load or undetectable status
• Presence of visible blood, genital ulcers, inflammation or other mucosal disruption that can increase transmission efficiency

Each of these factors is represented by a parameterised multiplier that scales the baseline act probability downward or upward within ranges supported by clinical trial data, observational studies and guideline estimates. Where published literature presents a range of plausible effect sizes, the engine selects values that are conservative from a risk estimation perspective rather than optimistically low.

4. Deriving the Posterior Event Probability

In the final stage, the model combines the partner prior, the act specific likelihood and the modifier terms into a single scalar quantity. At a high level, the engine computes:

• The probability that the partner is living with HIV, after adjusting for context (prior)
• The probability of transmission given HIV positive status and the specified act (likelihood)
• The net effect of all prevention and amplifying conditions (modifier product)

These components are combined multiplicatively to yield a posterior event level probability that represents the estimated chance of HIV acquisition from the described encounter. This value is specific to the parameter configuration implied by the user’s answers and is not simply read from a static table.

For communication purposes, the posterior probability is then expressed in two formats:

• A percentage, typically shown to four decimal places, which preserves the numerical precision of the estimate
• A “1 in X” representation that maps the same probability to an intuitive frequency format

The narrative report that accompanies this output describes the main drivers of the estimate for that encounter, identifies which inputs reduced or increased risk and links the probability to recommended testing windows and standard clinical guidance.