The author was invited to reply to this correspondence and declined the invitation.
Longitudinal analysis of histologic high-grade disease after negative cervical cytology according to endocervical status †
Article first published online: 30 JUL 2002
Copyright © 2002 American Cancer Society
Volume 96, Issue 5, pages 316–318, 25 October 2002
How to Cite
Vollmer, R. T. (2002), Longitudinal analysis of histologic high-grade disease after negative cervical cytology according to endocervical status . Cancer, 96: 316–318. doi: 10.1002/cncr.10713
- Issue published online: 11 OCT 2002
- Article first published online: 30 JUL 2002
In her recent article, Dr. Mitchell demonstrated that the absence of endocervical cells (ECs) in a Papanicolaou (PAP) smear did not indicate a subsequently increased rate of histologic high-grade abnormalities,1 and then concluded no justification exists for the early repeat testing of women whose Pap smears are negative but lack an endocervical component. In her article Dr. Mitchell discussed the presence of ECs in two sequential PAP smears obtained within 36 months of one another. She defined four cohorts: A, B, C, and D. Cohort A had ECs present in the first and second PAP smears. Cohort B had ECs present in the second PAP smear only. Cohort C had ECs present in the first PAP smear only, and Cohort D failed to demonstrate ECs in either PAP smear. Clearly, the comparison of Cohorts A and B is important because both demonstrated ample ECs in the second PAP smear and yet were not found to have different rates of subsequent histologic high-grade lesions. Thus, Dr. Mitchell's recommendation against early repeat testing appears a reasonable conclusion to draw from a study of just these two cohorts. Conversely, the data from the study provide further information regarding the relation between ECs and outcome and may help reconcile the disparate results between cross-sectional studies and longitudinal studies with regard to the significance of ECs.
The very low P values found in Tables 2, 3, and 4 in the study demonstrate that the presence of ECs was related significantly to all the outcomes.1 Thus, despite her final conclusion, nothing in Dr. Mitchell's study suggested that ECs were not important to the outcomes. Conversely, it appears unreasonable to assume these results occurred because collected ECs are a risk factor for cervical pathology. Thus, the questions to consider are how and why did the ECs relate to the outcomes. To study these issues, I applied a logistic regression model to the data in Tables 2, 3, and 4 from Dr. Mitchell's article.1, 2 If we symbolize the probability of observing an outcome of significant cervical pathology within 6 months of the second PAP smear as P, then the logistic analysis begins by modeling the logit transform of P as:
in which EC1 and EC2 are dummy variables and a and b are regression coefficients. EC1 is 1 if ECs are present in the first PAP smear; otherwise, EC1 is 0. Similarly, EC2 is 1 if ECs are present in the second PAP smear; otherwise, EC2 is 0. This approach allows one to evaluate how outcomes depend on the separate contributions of ECs in the first and second PAP smears. Table 1 provides the results for four of Dr. Mitchell's outcomes, and although in general these results are the same as the ones she obtained, they differ in how they emphasize the way in which the outcomes depended on EC1 and EC2.
|Outcome of histologic high-grade lesion|
|EC2||1.45||3.9 × 10−15|
|Outcome of histologic low-grade lesion|
|Outcome of any histologic lesion|
|Outcome of cytologic high-grade, or possible high-grade, lesion|
The results of Table 1 demonstrate two important findings. First, the six low P values indicate that in this longitudinal study, the outcomes depended significantly on the presence of ECs in PAP smears for at least six categories. The positive values for the coefficients in these six circumstances imply that there was greater detection of outcomes when ECs were present. Second, the larger coefficient values and lower P values for EC2 compared with EC1 indicate that the absence of ECs in the first PAP smear had less impact on a late outcome than the absence of ECs in the PAP smear obtained immediately prior to assessing the outcome. In other words, the presence of ECs is a time-dependent variable that has a lower impact on outcome as the distance in time from the PAP smear to measured outcome lengthens. The data in Dr. Mitchell's study also indicate that nearly 70% of those patients without ECs in the first PAP smear had ECs in the second PAP smear, suggesting that as time progressed a remote PAP smear without ECs became less important. Such time dependency fits with observations regarding the regression of cervical lesions,3 and it also relates to how exposure to different types of human papillomavirus may change with time. The time dependency of ECs in PAP smears also helps us understand how cross-sectional studies and longitudinal studies can obtain different results. Whereas the cross-sectional studies have examined the relation between ECs and an immediate outcome and found ECs to be significant, longitudinal studies have emphasized the presence of ECs in PAP smears long before the outcome is assessed. Just as in cross-sectional studies, the results of the study by Mitchell continue to emphasize how the presence of ECs in the most recent PAP smear is related to cervical pathology assessed near the time of that PAP smear. One explanation that fits our experience and intuition is that PAP smears without ECs result in lower sensitivity for detecting pathology. In other words, PAP smears without ECs are somehow less connected to an immediate sequence of events, including colposcopy and biopsy, which then lead to an immediate diagnosis of cervical pathology.
To illustrate, let us assume that there was no bias in how patients appeared in each of the four cohorts in Dr. Mitchell's study. In other words, at the outset each cohort should have had the same probability of having a histologic high-grade lesion at the end of the study. The logistic fit of the data suggests that (when ECs are present in the second PAP smear) this probability should be approximately 0.004. In Table 2 I compare the observed number of cases with histologic high-grade lesions with those predicted by this probability, and it can be seen that Cohorts C and D had far fewer high-grade lesions detected than would be expected had ECs been present in the second PAP smear. These results flow directly from the logistic model and Dr. Mitchell's collected data and in fact do not contradict her main conclusion that the status of ECs in a remotely obtained PAP smear do not relate significantly to high-grade histologic lesions observed much later. In other words, the importance of ECs appears to be limited largely to the most recent PAP smear, and perhaps we should be concerned only when we are dealing with the very first PAP smear in a woman or when the time since a PAP smear with ECs present has been long. At the very least, Dr. Mitchell's study data offer a way to reconcile cross-sectional studies with longitudinal ones, and it appears to raise the above hypothesis.
- 2Applied logistic regression. New York: John Wiley & Sons, Inc., 1989., .