David P Wilson1. 1. National Centre in HIV Epidemiology and Clinical Research, The University of New South Wales, Sydney, Australia. dwilson@nchecr.unsw.edu.au
Abstract
BACKGROUND: Exposures to sexually transmitted infections are discrete identifiable events. Interventions to prevent sexually transmitted infections have a certain probability of effectiveness in reducing risk in any given event. PURPOSE: Randomized control trials for sexually transmitted infections interventions are designed to estimate the effectiveness in preventing acquisition of infection. Typically, randomized control trials are run over a specific period of time and incidence in the control arm is compared with incidence in an intervention arm. However, it is possible that the effectiveness of an intervention over the duration of a clinical trial may be different to the actual effectiveness of the intervention in every single exposure event or the overall effectiveness over different periods than the duration of the trial. METHODS: In this study a simple mathematical framework is used, similar to methods in conception research, to describe the expected effectiveness that would be observed in a clinical trial of an intervention per-exposure and for clinically relevant shorter and longer durations than the trial, where each subject has multiple risk exposures. RESULTS: It is theoretically demonstrated that the actual effectiveness of the intervention per risk event is not equal to the overall preventative effectiveness of the intervention in preventing transmission over many exposures. Examples are given for sexually transmitted infections with diverse transmission probabilities (HIV and HPV) and for interventions with different levels of effectiveness (condoms and circumcision). The observed effectiveness of an intervention is likely to be maintained over many exposures for infections with low transmission risk (like HIV) but the observed effectiveness decreases substantially with number of exposures for moderate or high risk infections (like HPV). An equation is provided for interpreting randomized control trials' estimates of effectiveness with respect to various degrees of risk exposure. LIMITATIONS: The difficulty in adjusting the interpretation of randomized control trials results in this manner is that collection of accurate data on the number of discrete exposure events is not always possible and that there is substantial heterogeneity in degree of risk exposure between participants in trials. CONCLUSIONS: The implications of this analysis are that common interpretations of clinical trial interventions are insufficient for understanding the true efficacy of an intervention in some circumstances. Estimates of effectiveness in trials may misrepresent effectiveness per exposure event and effectiveness over a lifetime of risk. Care should be taken when designing protocols for analysis of trial results when the expected incidence is high. No change to the current practice of designing randomized control trials is suggested but analysis of trial data could be extended to calculate other statistics of effectiveness. A type of extrapolation and interpolation method for estimating levels of effectiveness is proposed.
BACKGROUND: Exposures to sexually transmitted infections are discrete identifiable events. Interventions to prevent sexually transmitted infections have a certain probability of effectiveness in reducing risk in any given event. PURPOSE: Randomized control trials for sexually transmitted infections interventions are designed to estimate the effectiveness in preventing acquisition of infection. Typically, randomized control trials are run over a specific period of time and incidence in the control arm is compared with incidence in an intervention arm. However, it is possible that the effectiveness of an intervention over the duration of a clinical trial may be different to the actual effectiveness of the intervention in every single exposure event or the overall effectiveness over different periods than the duration of the trial. METHODS: In this study a simple mathematical framework is used, similar to methods in conception research, to describe the expected effectiveness that would be observed in a clinical trial of an intervention per-exposure and for clinically relevant shorter and longer durations than the trial, where each subject has multiple risk exposures. RESULTS: It is theoretically demonstrated that the actual effectiveness of the intervention per risk event is not equal to the overall preventative effectiveness of the intervention in preventing transmission over many exposures. Examples are given for sexually transmitted infections with diverse transmission probabilities (HIV and HPV) and for interventions with different levels of effectiveness (condoms and circumcision). The observed effectiveness of an intervention is likely to be maintained over many exposures for infections with low transmission risk (like HIV) but the observed effectiveness decreases substantially with number of exposures for moderate or high risk infections (like HPV). An equation is provided for interpreting randomized control trials' estimates of effectiveness with respect to various degrees of risk exposure. LIMITATIONS: The difficulty in adjusting the interpretation of randomized control trials results in this manner is that collection of accurate data on the number of discrete exposure events is not always possible and that there is substantial heterogeneity in degree of risk exposure between participants in trials. CONCLUSIONS: The implications of this analysis are that common interpretations of clinical trial interventions are insufficient for understanding the true efficacy of an intervention in some circumstances. Estimates of effectiveness in trials may misrepresent effectiveness per exposure event and effectiveness over a lifetime of risk. Care should be taken when designing protocols for analysis of trial results when the expected incidence is high. No change to the current practice of designing randomized control trials is suggested but analysis of trial data could be extended to calculate other statistics of effectiveness. A type of extrapolation and interpolation method for estimating levels of effectiveness is proposed.