Could misreporting of condom use explain the observed association between injectable hormonal contraceptives and HIV acquisition risk?

Abstract

Objective: Some observational studies have suggested an association between the use of hormonal contraceptives (HC) and HIV acquisition. One major concern is that differential misreporting of sexual behavior between HC users and nonusers may generate artificially inflated risk estimates.

Study design: We developed an individual-based model that simulates the South African HIV serodiscordant couples analyzed for HC-HIV risk by Heffron et al. (2012). We varied the pattern of misreporting condom use between HC users and nonusers and reproduced the trial data under the assumption that HC use is not associated with HIV risk. The simulated data were analyzed using Cox proportional hazards models, adjusting for the reported level of condom use.

Results: If HC users overreport condom use more than nonusers, an apparent excess risk could be observed even without any biological effect of HC on HIV acquisition. With 45% overreporting by HC users (i.e., 9 out of every 20 sex acts reported with condoms are actually unprotected) and accurate condom reporting by nonusers, a true null effect can be inflated to give an observed hazard ratio (HR̂) of 2.0. In a different population with lower overall reported condom use, artificially high HR̂s can only be generated if non-HC users underreport condom use.

Conclusion: Differential condom misreporting can theoretically produce inflated HR̂ values for an association between HC and HIV even without a true association. However, to produce a doubling of HIV risk that is entirely spurious requires substantially different levels of misreporting among HC users and nonusers, which may be unrealistic.

Implications: Considerably differential amounts of condom use misreporting by HC users and nonusers would be needed to produce entirely spurious observed levels of excess HIV acquisition risk among HC users when there is actually no true association.

Keywords: Condom use; DMPA; HIV; Injectable hormonal contraceptives; Mathematical modelling; Misreporting bias.

Fig. 1

Schematic of condom use and condom reporting in the model. No misreporting means that all sex acts are reported accurately with respect to condom use. −100% means that all sex acts using condoms are reported as unprotected (underreporting); 100% means that all unprotected sex acts are reported with condom use (overreporting). In this example, with 70% overreporting, an individual who reported using condoms in 10 out of 10 sex acts would be assumed to actually use a condom in only 3 sex acts (70% of their reported sex acts with condom use are reassigned to unprotected sex acts). With 60% underreporting (or −60% misreporting), an individual who reported using condoms in 0 sex act out of 10 would be assumed to actually use a condom in 6 (60% of their unprotected sex acts are reassigned as using condoms).

Fig. 2

Model-generated HRs with different levels of reported condom use and underlying true HR. Point estimates of HRs observed in the simulated data under different levels of misreporting condom use among HC users (15% of women; vertical axis) and nonusers (85% of women; horizontal axis). (a) Scenario 1: reported condom use from the Partners in Prevention HSV/HIV Transmission Study, HR=1.0. Ninety-one percent of both HC users and nonusers report consistent condom use. Specific misreporting biases are labeled 1–6 which are discussed in the text and plotted in Fig. 2d (dark gray). These are (1) no condom misreporting in either group; (2) 45% overreporting in both HC users and nonusers; (3) 45% overreporting in HC users only; (4) 80% overreporting in HC users only; (5) 45% overreporting in HC users and 20% in nonusers and (6) 30% overreporting in HC users and 30% underreporting in nonusers. A smoothed surface was fitted to the geometric means of the model simulations using LOESS (R²=0.9956). (b) Scenario 2: reported condom use from the Partners in Prevention HSV/HIV Transmission Study, HR=1.2. Ninety-one percent of both HC users and nonusers report consistent condom use. Specific misreporting biases are labeled 1 and 2: (1) 30% overreporting in HC users only and (2) 40% overreporting in HC users and 10% in nonusers. Surface fitted using LOESS (R²=0.9955). c. Scenario 3: 50% condom use, HR=1. Fifty percent of both HC users and nonusers report consistent condom use. Specific points labeled 1–6 refer to the same misreporting behaviors detailed in Fig. 2a and are plotted in Fig. 2d (light gray). Surface fitted using LOESS (R²=0.9965). (d) Point estimates from Scenarios 1 and 3. Point estimates of HRs observed in the simulated data from Scenarios 1 (dark gray, Fig. 2a) and 3 (light gray, Fig. 2c). Points 1–6 refer to the specific misreporting biases identified in Fig. 2a. Error bars represent 90% of the variation in the model-generated $\widehat{\mathrm{HR}}$s at that point, and the dashed black line marks the HR for all HC observed in Heffron et al. (HR=1.98 [1.06–3.68]) .