A New Window into the Natural History of Human Papillomavirus Infection: A View from the ALTS (Atypical Squamous Cells of Undetermined Significance/Low‐Grade Squamous Intraepithelial Lesions Triage Study) Trial

Persistent cervical infection with any of ∼15 oncogenic types of human papillomavirus (HPV) places a woman at a substantially high risk of developing high-grade precancerous lesions and, subsequently, if the lesions are left untreated, of invasive cervical cancer [1, 2, 3, 4]. Most cases eventually arise after persistent infection with HPV types 16 or 18 [5, 6, 7], the 2 types targeted by a highly efficacious HPV vaccine that has been recently approved [8]. Proper understanding of the duration of HPV infections and their likelihood to evolve into cervical precancer has been the focus of many cohort studies [9, 10, 11, 12, 13, 14, 15, 16]. The research question is thus not trivial; although HPV infections are common, most clear spontaneously, particularly among young women [9, 15]. Preventing persistent HPV infections via vaccination or detecting them via HPV testing in the context of opportunistic or organized screening are now the new frontiers of cervical cancer prevention. Molecular epidemiologic studies that have contributed to our understanding of the dynamics of HPV persistence have resorted mostly to frequentist statistical approaches to assessing the duration of infections and their likelihood to clear or persist. Use of actuarial techniques to account for censoring during follow-up of female subjects (i.e., absence of data that would have permitted fully accounting for the beginning and end of an infection episode) and time-to-event regression modeling techniques that account for covariate effects have been the methodological mainstays of such studies. The median time to clearance (i.e., the time point when 50% of those originally infected no longer retain their infections) varies from 8 to 17 months for the oncogenic types and from 4 to 15 months for the nononcogenic types (reviewed in [17]). Contributing to the heterogeneity among findings are differences across studies in the target populations, in the time intervals used to collect cervical samples, in the performance of HPV testing methods, and in whether or not prevalent infections were included in the calculations, among other study design and computational differences. Most studies, however, have focused on asymptomatic women, mostly without cervical cytological abnormalities. In other words, the window into the natural history of HPV infection that is “visualized” presumably reflects events happening more upstream during the process that begins with infection and evolves into lesions. Enter the ALTS (atypical squamous cells of undetermined significance [ASCUS]/ low-grade squamous intraepithelial lesions [LSIL] Triage Study) trial, a randomized controlled trial of management options for women diagnosed with ASCUS (equivocal atypias) or LSIL. Findings from the ALTS trial have already changed the landscape of medical practice regarding follow-up of abnormal Pap smears [18, 19, 20]. In this issue of the Journal, Plummer et al. [21] have used data from the enrollment and follow-up phases of the ALTS trial involving >5000 women across the United States to provide an in-depth assessment of the time dynamics of prevalent and incident HPV infections detected during scheduled 6-month visits over 24 months. In addition to the sheer size and rigor with which the study was conducted, this report breaks new ground by using a scientifically cogent Markov model that takes the unified approach of fitting the entire dynamics of infection, including prevalence at enrollment, acquisition, and clearance, as connected processes. The model considers infection by different HPV types as independent events, but, in separate analyses, the authors extended the basic model to examine the risk of being positive for a given type at the next visit conditioned on having another HPV type at the current visit. This permitted them to assess interaction among types. However, because acquisition of HPV infection regardless of type is dependent on shared behavioral, environmental, or host-specific risk factors that could not be entirely subsumed in the covariate information available in the ALTS data set (age, study center and arm, referral cytological diagnosis, and number of male sex partners since the last visit), the above model was further extended to accommodate individual random variation in how infections by type tended to co-occur in ALTS. Their model used a Bayesian framework for the probabilistic statements and stochastic simulation of the distribution of all parameters. It is worth noting that the computations required to generate the tables and figures in the article by Plummer et al. [21] would have been a daunting proposition outside of the realm of military or engineering research centers just 10 years ago. Recent advances in software design for Bayesian analysis and Monte Carlo simulation, coupled with the availability of shared high-performance computing for research, made the authors' ambitious unifying model a realistic undertaking. What are the main lessons learned by Plummer et al.? One of the most salient findings was the conclusion that prevalent infections at enrollment clear relatively fast with 80% and 90% of them having cleared after 12 and 24 months, respectively, as average estimates across all HPV types. The median time to clearance ranged from a little over 4 months for HPV-56 to nearly 8 months for HPV-62, with an average across all types of ∼5 months. Contrasting these estimates with the ones mentioned above for studies conducted mostly among cytologically negative women suggests a paradoxical situation, because ALTS trial estimates are noticeably shorter or are at the low end of the range of median times from previous investigations. Yet, by design, all ALTS participants had an underlying equivocal or minor-grade abnormality that would place the aforementioned window into the natural history of HPV infection peeking at events happening more downstream and closer to lesion development. It is thus plausible to assume that considerable selection would have already happened when one examines the dynamics of an infection after it has colonized the cervix long enough to reach a viral load that is sufficient to disrupt the maturation of the epithelium and make cytological abnormalities recognizable on a Pap smear. Is it possible that the stopwatch had already started for all prevalent infections in ALTS long before the first HPV-positive result was documented? Could lack of controlling for this left-censoring situation have led to an underestimation of the average duration of an infection? The authors were mindful of the problem and also assessed the duration of incident infections, which do not pose the left-censoring problem because their onset happened during follow-up. Although it is semantic license to talk about “incident” infections in women who have already accrued substantial opportunity for HPV exposure, the exercise was nonetheless useful. What they learned was that the behavior of prevalent and incident infections was very similar in terms of how fast they cleared. We are therefore left with the seemingly counterintuitive finding that HPV infections in women with minor squamous cytological abnormalities clear just as fast (if not faster) than those in women whose infections coexist with negative Pap smears. Another important finding by Plummer et al. [21] was the realization that the longer an infection persists, the more likely it is to continue to be detectable. Averaging over all types, a prevalent infection that lasted to month 6 was nearly twice as likely to last to month 24 if it continued to be detected at month 18, with comparable results for incident infections [21]. This conclusion also has a counterintuitive ring to it. We know from this study and previous ones that most infection episodes clear eventually and that any given investigation collects for each woman a string of HPV test results (e.g., ++---, -+++-, -++--, +++++, etc.). Gaps in positivity and considerations about censoring aside, if one intuitively looks at any 2 consecutive time points, such as 2 sampling observations 6 months apart, the probability that the second HPV result will be negative increases as one gets to the temporal end of a string of plus signs. The exponential decay–like curve of most infection episodes suggests an underlying skewed distribution of duration times. Therefore, the expectation is that the closer an observer gets to the end of a string of plus signs, the more likely that it is about to end. How can one reconcile the study's findings with the practical expectations described above for the duration of infection episodes and for likelihood of persistence? The ALTS's findings are most representative of what can be expected for women who are candidates to colposcopy because of an LSIL smear or because of an HPV-positive ASCUS. These women have elevated risks of cervical intraepithelial neoplasia (CIN) of grades 2 and 3, which were common findings during follow-up in ALTS. Could excessive right-censoring because of excisional therapy after a histological diagnosis of CIN2 or worse have biased the estimates of infection duration? The authors excluded from the analysis 542 women with a diagnosis of CIN3 or worse because they could have been prevalent at enrollment but left those with CIN2 (presumably a higher number) because of the uncertainties associated with distinguishing these lesions from HPV-attributable effects. Although a sensitivity analysis indicated that the results would not have materially changed had CIN3 cases not been excluded, it is possible that the availability of HPV test results may have been variable among women with a CIN2 diagnosis at the diagnostic visit and subsequently. Notwithstanding these caveats and others related to trade-offs needed to take advantage of the sophistication of the modeling technique, the findings from the ALTS trial are not counterintuitive if one considers a biological model in which the most critical carcinogenic events tend to follow shortly after an incident HPV infection with an oncogenic type. Under this assumption, a sample of patients with ASCUS and LSIL would not be much different from one of comparable age selected randomly among women attending screening or taken from the general population. Recent corroborating data from the placebo arm of HPV vaccine trials [22] or from follow-up of young women [23] indicate that many CIN events are detectable within a few months after the incident HPV infections that presumably caused them. Finally, another important conclusion by Plummer et al. [21] was the virtual noninteraction among HPV types in how they behave as independent infection episodes. This is not to say that risk of acquiring infections by different types and how long they take to clear are totally independent events, as one would, for example, consider any 2 HPV types as distinct as an episode of influenza can be from one of urinary tract infection. Undeniably, infections by different HPV types are acquired by the same route (i.e., sexual activity), with host-related and other factors contributing ancillary empirical value in explaining the codistribution of HPV types in the population [13]. The merit of the modeling approach used by the authors was in looking beyond this source of variability, both measured (sociodemographic and behavioral data collected in ALTS) and unmeasured (via inclusion of “frailty” terms). In other words, they assessed the influence of finding a given type on the occurrence of another one over and above what can be expected from factoring in the shared components of risk and susceptibility. This conclusion goes a long way in assuaging concerns that HPV vaccination could lead to type replacement and a theoretical concern that eliminating HPV types 16 and 18 would open ecologic niches for other oncogenic types to occupy.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.