Missing SOGI Data Blunt PrEP Disparity Detection in a Safety-Net System
Federal reporting changes in 2025 raised a practical surveillance problem with immediate consequences for HIV prevention: when sexual orientation and gender identity data disappear from routine records, health systems lose the ability to see who is and is not receiving HIV preexposure prophylaxis orders at expected rates. This cross-sectional analysis from a large safety-net health system in Southern California tested exactly how much signal is lost when SOGI data are complete, partially missing, or reduced to sex assigned at birth alone. The answer was blunt. Complete SOGI data allowed disparity detection; SAAB-only surveillance did not.
The cohort was large, with 165,198 adults and a mean age of 45.0 years. Most patients, 97,621, were female sex assigned at birth, while 5,790 patients were lesbian, gay, bisexual, or another sexual minority, and 303 were transgender or gender diverse. Only 1,406 patients had a documented PrEP order, a reminder that even in a safety-net setting with access to electronic records, the prevention signal is sparse relative to the size of the denominator. Sparse outcomes make surveillance fragile when key identity fields are incomplete.
Complete SOGI Data Exposed Differences That SAAB Alone Concealed
Under the complete SOGI analytic scenario, the adjusted probability of a documented PrEP order among cisgender patients assigned male at birth was 26.7 percentage points higher for LGB+ patients than for heterosexual patients, with a 95% confidence interval of 24.9 to 28.5 percentage points and a P value below .001. Among all patients assigned male at birth, the adjusted probability was 12.3 percentage points higher for transgender and gender diverse patients than for cisgender patients, with a 95% confidence interval of 7.0 to 17.6 percentage points and a P value below .001. Neither difference was significant among patients assigned female at birth.
One sentence captures the operational problem: when the record only knows sex assigned at birth, the disparity signal collapses into a crude male-versus-female comparison.
Under SAAB-only surveillance, the only measurable difference was between male and female sex assigned at birth, a 1.4 percentage-point difference with a 95% confidence interval of 1.3 to 1.5 percentage points and P below .001. SOGI-specific differences could not be estimated at all. In clinical terms, that means the system could count sex but not identity or orientation, and therefore could not identify where prevention gaps were concentrated. For PrEP programs, that is more than a statistical inconvenience. It shifts the lens away from populations with the clearest prevention need and toward a metric that is too blunt to guide equity-focused outreach.
Why Identity Fields Matter for HIV Prevention Surveillance
PrEP allocation depends on recognizing risk, and risk in HIV prevention is not captured by sex assigned at birth alone. Sexual minority and transgender populations may have distinct patterns of healthcare engagement, stigma exposure, and risk recognition in clinical encounters. A surveillance system that strips out those fields can still describe the patient population in aggregate, but it cannot separate whether low PrEP ordering reflects low need, poor access, missed documentation, or clinician under-recognition of risk. The biology of HIV acquisition is unchanged by the reporting framework; what changes is the health system’s ability to identify who should be offered prevention before exposure occurs.
The design of this study made the stakes visible. Investigators used electronic health record data from patients 18 years or older who were registered as of October 28, 2025 and had documented PrEP orders between January 2, 2019 and October 27, 2025. Registration dates were unavailable, which prevented conventional follow-up analysis, so the study focused on cross-sectional surveillance scenarios instead. Logistic regression models adjusted for age and race and ethnicity and included interactions between sexual orientation or gender identity and sex assigned at birth. Marginally standardized probabilities and adjusted percentage-point differences were estimated within SAAB strata, and each nonzero missingness level was tested through 250 simulations. That simulation framework mattered because it allowed the study to ask a hard question: how much missingness can a surveillance system tolerate before disparities become statistically invisible?
Partial Collection Preserved Some Estimability, but Precision Eroded Quickly
Complete removal was not the only threat. Partial missingness changed the quality of inference in a graded way. Under joint missing-completely-at-random deletion, positive-direction detectability of the male-SAAB sexual orientation difference stayed at 100% through 90% missingness. The gender identity contrast was less durable. Detectability fell from 100% at 40% missingness to 51.6% at 80% missingness and 19.2% at 90% missingness.
Those numbers translate into a simple surveillance message: a system can lose a large fraction of SOGI data and still estimate some disparities, but the precision needed to trust gender identity findings drops much sooner than many administrators would expect. In practice, a positive association can survive missingness longer than the narrower transgender and gender-diverse signal, which is consistent with the much smaller size of the TGD subgroup in the cohort. Small subgroups are the first to disappear into noise when field completion worsens.
Selective missingness caused further attenuation of the estimates. In other words, when missingness was not random, the observed disparities moved closer to null. This is the most consequential finding for real-world reporting systems, because missing data in clinical records are rarely neutral. Identity fields may be left blank more often in specific settings, patient populations, or workflows, and that pattern can distort exactly the groups most likely to experience barriers to PrEP access.
Surveillance Policy Has Direct Consequences at the Bedside
PrEP is not a theoretical equity measure. It is a documented order in a chart, and the choice to order it depends on whether clinicians know whom they are caring for and whether systems can flag prevention gaps accurately. When SOGI data are complete, disparity measurement is possible. When those fields are absent, the best available signal may be too coarse to reveal where prevention is being missed. The result is not simply less information; it is a narrower view of who receives HIV prevention and who does not.
This study also shows why partial data are preferable to no data, but only up to a point. Partial collection preserved estimability, yet precision steadily weakened as missingness increased. For transgender and gender diverse patients in particular, the loss of precision was steep. A reporting policy that removes SOGI fields entirely would not just reduce administrative complexity; it would erase the ability to detect inequities in one of the clearest biomedical prevention measures available for HIV.
The analysis stops short of proving why individual patients received or did not receive PrEP orders, and it does not measure actual medication use, adherence, or HIV incidence. It also relies on a single safety-net system in Southern California, so the magnitude of these effects may differ elsewhere. Even so, the central inference is difficult to avoid: without complete SOGI data, health systems may continue ordering PrEP, but they will be doing so with blind spots that make disparity detection far less reliable.