NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Kulasingam SL, Havrilesky L, Ghebre R, et al. Screening for Cervical Cancer: A Decision Analysis for the U.S. Preventive Services Task Force [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2011 May. (Evidence Syntheses, No. 86s.)

Cover of Screening for Cervical Cancer: A Decision Analysis for the U.S. Preventive Services Task Force

Screening for Cervical Cancer: A Decision Analysis for the U.S. Preventive Services Task Force [Internet].

Show details

Appendix BModel Description

An extensive description of the structure of the model, including the natural history and screening components, is published elsewhere.9,17 A summary of the model is provided here, and key inputs to the model are summarized in Appendix B Tables 1 and 2.

Appendix B Table 1. Estimates of Incidence, Progression, and Regression Applied to HPV and CIN States in Markov Model.

Appendix B Table 1

Estimates of Incidence, Progression, and Regression Applied to HPV and CIN States in Markov Model.

Appendix B Table 2. Estimates of Incidence, Progression, and Regression Applied to HPV and CIN States in Markov Model.

Appendix B Table 2

Estimates of Incidence, Progression, and Regression Applied to HPV and CIN States in Markov Model.

Structure of the Model

The model has two components. The first component is a 20-state Markov model66 that simulates the natural history of cervical cancer in the absence of screening. The second component is an intervention model that represents possible screening strategies. The model was originally developed using DATA 3.0 software (TreeAge Software, Inc., Boston, MA); updates were made using TreeAge Pro 2010, HealthCare Version (TreeAge Software, Inc., Boston, MA).

Natural History

The model follows a cohort of women from age 12 to 100 years and assumes that, at the beginning of the simulation, no one is infected with HPV or CIN1, CIN2-3, or cancer. Cycle lengths are 1 year long. Each year women can be infected with HPV. Women infected with HPV can undergo regression, no change, or progression to CIN. Although most progress from HPV to CIN1, a proportion progresses directly to CIN2-3. Women with CIN1 can undergo regression (to either “well” or the HPV-infected state), no change, or progression to CIN2-3. Women with CIN2-3 can regress, stay in the same state, or progress to Stage I cancer. Women with cancer either become symptomatic or progress through Stages II-IV. Once a cancer diagnosis is made, the probability of survival is stage-specific. Women without cancer are at risk for hysterectomy for other causes, and all women are at risk for death from other causes. The states and allowed transitions of the natural history model are summarized in Appendix B Figure 1.

This figure shows how women in the cohort flow between different states. The states are shown in ovals on two separate rows. The first row of states, from left to right, are as follows: normal, HPV, CIN 1, CIN 2 to 3, and cancer stages I to IV. The bottom row of states, from left to right, are as follows: dead, hysterectomy, and cancer death. Arrows show the movement that is allowed from one state to another and appear in the figure as follows: arrows from normal to HPV, dead, and hysterectomy; arrows from HPV to normal, dead, CIN 1, and hysterectomy; arrows from CIN 1 to HPV, dead, hysterectomy, and CIN 2 to 3; arrows from CIN 2 to 3 to CIN 1, dead, hysterectomy, and cancer stages I to IV; and arrows from cancer stages I to IV to cancer death and death.

Appendix B Figure 1

Disease States and Allowed Transitions for the Natural History Component of the Cervical Cancer Markov Model.

Interventions

Screening Strategies

Details and assumptions for the different screening strategies modeled are presented in the Methods section.

Diagnostic Strategies for Abnormal Pap and HPV Test Results

Strategies for the followup of abnormal HPV and cytology test results are based on the 2006 American Society for Colposcopy and Cervical Pathology consensus guidelines for the management of abnormal screening tests and CIN.33

Measures of Effectiveness

The model is used to estimate the number of false-positives, colposcopies performed, CIN2-3 cases detected, cancer cases detected, and cancer deaths. The main outcome is colposcopies per life-year. In addition, average lifetime costs, life expectancy, and quality-adjusted life expectancy are estimated. Incremental ratios of the difference in colposcopies divided by the difference in life expectancy were calculated in order to determine which strategies should be considered for the recommendation update.

Assumptions

Population

The model follows a cohort of U.S. women from age 12 to 100 years.

Histological Subtypes of Invasive Cervical Cancer

Squamous cancer of the cervix accounts for approximately 80 to 85 percent of invasive cervical cancer cases. Adenocarcinoma, which accounts for another 10 to 15 percent, may be increasing in incidence.67 Cervical cytology may also be less sensitive for adenocarcinoma. However, we did not distinguish between histologic subtypes in any of the estimates for screening or treatment. This is consistent with the approach taken in the original model and also consistent with other models.

Patient and Provider Behavior

Consistent with other models, this model assumes that all women in the cohort (100 percent) receive the screening test at the appropriate interval and that all patients receive appropriate diagnostic and therapeutic interventions based on the results of the screening tests for the base-case analysis. The implications of less than perfect adherence to screening are explored in sensitivity analyses.

Parameter Estimates From Available Data

Hysterectomy for benign disease. Age-specific hysterectomy rates are based on estimates from Keshavarez et al.68

Incidence of HPV infection. Since estimates of test performance for HPV DNA testing are conditioned on underlying histology rather than HPV type, no distinction is made between different types of HPV. The incidence, progression, and regression estimates are averages for all viral types. The age-specific estimates for HPV incidence in the model were back-calculated in order to produce an HPV prevalence curve that is consistent with the reported literature (Appendix B Table 1 and Appendix B Figure 2).13-15,69 In particular, the prevalence curve shows a peak in prevalence and magnitude that is similar to that reported in U.S. population-based studies by Dunne et al13 and Kulasingam et al.14

This figure shows the model–predicted age-specific prevalence of HPV and observed prevalence of HPV based on three studies. The x-axis shows age, grouped in increments of 5 years and varying from 15 to 19 to 80 to 84 years of age. The y axis shows the prevalence of HPV and varies from 0 to 0.35. Bars show the model–predicted age-specific prevalence of HPV. The model-predicted prevalence is 0.15 for ages 15 to 19 years, peaks at 0.24 for ages 20 to 24, and then decreases to 0.03 for ages 80 to 84.

Appendix B Figure 2

Duke Cervical Cancer Model (Main Analysis): Prevalence of HPV*. *Dunne prevalence estimates are measured in 10-year increments beginning at age 30 years (30-39, 40-49, and 50-59). Kulasingam estimates are measured in 5-year increments up to age 34 years, (more...)

Regression, persistence, and progression of HPV infection. These estimates are averaged for all types and are primarily based on the older model but confirmed with more recent studies. The one significant change is the estimate of progression from CIN2-3 to cancer. In the older model, this was estimated to be approximately 4 percent per year. In this model, we have revised the estimate for younger women (aged 30 years and younger) to reflect recent analyses that show that progression from CIN3 to cancer is approximately 1 percent per year.25-26

Revised natural history model to account for different progression and regression rates of CIN. Estimates for progression and regression between low-grade and high-grade neoplasia are from the original model as well as an updated review of the literature. Historically, CIN has been viewed as a continuum, with progression from HPV infection to CIN1, CIN2, and CIN 3 assumed to take place over a period of decades, representing a slow progression of disease. The original model was developed to represent this view of CIN. Recently, however, studies suggest that CIN1 and CIN2-3 may be established separately, and that young women can develop a CIN2-3 lesion within a short period of time (2 years).23,70-72 Based on these studies as well as others, Baseman and Koutsky have proposed a revised view of CIN, with an early establishment of high-grade lesions, but the majority regressing, with only a minority progressing.24 However, since it is unclear whether this view of the natural history is applicable to all women, we developed a model that reflects a higher burden of disease in young women in particular, but with most of the disease regressing, and only a small proportion progressing per year. Details of the estimates used are presented in Appendix B Table 2. The model was calibrated to produce an HPV prevalence curve and cancer incidence and mortality curve similar to those observed in large screening studies and SEER data (Appendix B Figures 3-7). Assuming that most women undergo screening at least every 3 years, the model predicts a lifetime risk of developing cancer of 0.63 and a lifetime risk of dying from cancer of 0.17, compared to the SEER estimates of 0.672 and 0.23, respectively.58

This figure shows the model-predicted and SEER observed age-specific cancer incidence. Age in years is presented in 10-year increments from age 10 to age 90 along the x-axis. Incident cancer cases per 100,000 women is presented ranging from 0 to 90 along the y-axis.

Appendix B Figure 3

Duke Cervical Cancer Model: SEER Age-Specific Cancer Incidence Assuming No Screening, Screening Every 1 Year, or Screening Every 3 Years.

This figure shows the model-predicted and SEER observed age-specific cancer mortality. Age is presented in 10-year increments from age 10 to age 90 along the x-axis. Cancer mortality per 100,000 women is presented ranging from 0 to 30 along the y-axis.

Appendix B Figure 4

Duke Cervical Cancer Model: SEER Age-Specific Cancer Mortality Assuming No Screening, Screening Every 1 Year, or Screening Every 3 Years.

This figure shows the model-predicted age-specific prevalence of HPV and observed prevalence of HPV based on three studies. The x-axis shows age, grouped in increments of 5 years and varying from 15 to 19 to 80 to 84 years of age. The model-predicted prevalence is based on different values for the natural history parameters than those used to generate the model-predicted prevalence curve in Appendix B. Figure 2. The y-axis shows the prevalence of HPV and varies from 0 to 0.35. Bars show the model-predicted age-specific prevalence of HPV. The model-predicted prevalence is 0.15 for ages 15 to 19 years, peaks at 0.24 for ages 20 to 24, and then decreases to approximately 0.03 for ages 80 to 84.

Appendix B Figure 5

Prevalence of HPV-Revised Natural History Model (Sensitivity Analysis Only). *Dunne prevalence estimates are measured in 10-year increments beginning at age 30 years (30-39, 40-49, and 50-59). Kulasingam estimates are measured in 5-year increments up (more...)

This figure shows the model-predicted and SEER observed age-specific cancer incidence. Age is presented in 10-year increments from age 10 to age 90 along the x-axis. Cancer incidence per 100,000 women is presented ranging from 0 to 90 along the y-axis.

Appendix B Figure 6

Revised Natural History Model (Sensitivity Analysis Only): SEER Age-Specific Cancer Incidence Assuming No Screening, Screening Every 1 Year, or Screening Every 3 Years.

This figure shows the model-predicted and SEER observed age-specific cancer mortality. Age is presented in 10-year increments from age 10 to age 90 along the x-axis. Cancer mortality per 100,000 women is presented ranging from 0 to 30 along the y-axis.

Appendix B Figure 7

Revised Natural History Model (Sensitivity Analysis Only): SEER Age-Specific Cancer Mortality Assuming No Screening, Screening Every 1 Year, or Screening Every 3 Years.

Natural history of invasive cancer. Estimates of the progression rate and the likelihood of symptoms (since cases would only be detected upon presentation with symptoms) by stage are from the original model and presented in Appendix B Table 3. These estimates were used for both natural history models.

Appendix B Table 3. Estimates of Symptoms, Progression, and Survival Used for Invasive Cervical Cancer States in Markov Model.

Appendix B Table 3

Estimates of Symptoms, Progression, and Survival Used for Invasive Cervical Cancer States in Markov Model.

Stage-specific survival. Survival probabilities at 1, 2, 3, 4, and 5 years post-diagnosis for each stage are from SEER data.60 Five-year survival rates based on these data are: Stage I (local), 91.3 percent; Stages II-III (regional), 54 percent; and Stage IV (distant), 15.8 percent. An assumption is made that there is no cancer-related mortality after 5 years. This assumption is consistent with the original version of the model and also allows for comparison with other models. In a sensitivity analysis, ratios of relative survival for women aged 50 to 69 years and 70 years and older compared to ratios of overall survival were calculated to address the issue of decreased survival in these older age groups. These ratios were 0.97 (for women aged 50 to 69 years) and 0.93 (for women aged 70 years and older) for Stage I; 1.03 and 0.78, respectively, for Stage II-III; and 0.81 and 0.65, respectively, for Stage IV.

Non-cervical cancer mortality. Mortality from causes other than cervical cancer is estimated by subtracting age-specific cervical cancer mortality rates from age-specific all-cause mortality rates using U.S. life tables for women.78

PubReader format: click here to try

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (1.1M)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...